Content uploaded by Shridhar C. Ghagane
Author content
All content in this area was uploaded by Shridhar C. Ghagane on Jan 09, 2023
Content may be subject to copyright.
Lecture Notes in Networks and Systems 558
Kingsley A. Ogudo
Sanjoy Kumar Saha
Debnath Bhattacharyya Editors
Smart
Technologies
in Data
Science and
Communication
Proceedings of SMART-DSC 2022
Lecture Notes in Networks and Systems
Volume 558
Series Editor
Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences,
Warsaw, Poland
Advisory Editors
Fernando Gomide, Department of Computer Engineering and Automation—DCA,
School of Electrical and Computer Engineering—FEEC, University of
Campinas—UNICAMP, São Paulo, Brazil
Okyay Kaynak, Department of Electrical and Electronic Engineering,
Bogazici University, Istanbul, Turkey
Derong Liu, Department of Electrical and Computer Engineering, University of
Illinois at Chicago, Chicago, USA
Institute of Automation, Chinese Academy of Sciences, Beijing, China
Witold Pedrycz, Department of Electrical and Computer Engineering, University of
Alberta, Alberta, Canada
Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland
Marios M. Polycarpou, Department of Electrical and Computer Engineering,
KIOS Research Center for Intelligent Systems and Networks, University of Cyprus,
Nicosia, Cyprus
Imre J. Rudas, Óbuda University, Budapest, Hungary
Jun Wang, Department of Computer Science, City University of Hong Kong,
Kowloon, Hong Kong
The series “Lecture Notes in Networks and Systems” publishes the latest
developments in Networks and Systems—quickly, informally and with high quality.
Original research reported in proceedings and post-proceedings represents the core
of LNNS.
Volumes published in LNNS embrace all aspects and subfields of, as well as new
challenges in, Networks and Systems.
The series contains proceedings and edited volumes in systems and networks,
spanning the areas of Cyber-Physical Systems, Autonomous Systems, Sensor
Networks, Control Systems, Energy Systems, Automotive Systems, Biological
Systems, Vehicular Networking and Connected Vehicles, Aerospace Systems,
Automation, Manufacturing, Smart Grids, Nonlinear Systems, Power Systems,
Robotics, Social Systems, Economic Systems and other. Of particular value to
both the contributors and the readership are the short publication timeframe and
the world-wide distribution and exposure which enable both a wide and rapid
dissemination of research output.
The series covers the theory, applications, and perspectives on the state of the art
and future developments relevant to systems and networks, decision making, control,
complex processes and related areas, as embedded in the fields of interdisciplinary
and applied sciences, engineering, computer science, physics, economics, social, and
life sciences, as well as the paradigms and methodologies behind them.
Indexed by SCOPUS, INSPEC, WTI Frankfurt eG, zbMATH, SCImago.
All books published in the series are submitted for consideration in Web of Science.
For proposals from Asia please contact Aninda Bose (aninda.bose@springer.com).
Kingsley A. Ogudo · Sanjoy Kumar Saha ·
Debnath Bhattacharyya
Editors
Smart Technologies in Data
Science and Communication
Proceedings of SMART-DSC 2022
Editors
Kingsley A. Ogudo
University of Johannesburg
Johannesburg, South Africa
Debnath Bhattacharyya
Department of Computer Science
and Engineering
Koneru Lakshmaiah Education Foundation
Guntur, India
Sanjoy Kumar Saha
Department of Computer Science
and Engineering
Jadavpur University
Kolkata, West Bengal, India
ISSN 2367-3370 ISSN 2367-3389 (electronic)
Lecture Notes in Networks and Systems
ISBN 978-981-19-6879-2 ISBN 978-981-19-6880-8 (eBook)
https://doi.org/10.1007/978-981-19-6880-8
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Singapore Pte Ltd. 2023
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Conference Committee Members
Organizing Committee
General Chairs
Philippe Fournier-Viger, Shenzhen University, Guangdong, China
T. Pavan Kumar, Koneru Lakshmaiah Education Foundation, Vaddeswaram, Guntur,
Andhra Pradesh, India
Advisory Board
Yu-Chen Hu, Providence University, Taichung City, Taiwan
Zhihan Lv, Uppsala University, Sweden
Osvaldo Gervasi, University of Perugia, Italy
Paul S. Pang, Unitec Institute of Technology, New Zealand
Andrzej Goscinski, Deakin University, Australia
Jason Levy, University of Hawaii, Hawaii, USA
Tai-hoon Kim, Konkuk University, South Korea
Sabah Mohammed, Lakehead University, Ontario, Canada
Jinan Fiaidhi, Lakehead University, Ontario, Canada
Y. Byun, Jeju National University, South Korea
Amiya Bhaumick, LUC KL, Malaysia
Dipti Prasad Mukherjee, ISI Kolkata, India
Sanjoy Kumar Saha, Jadavpur University, Kolkata, India
Sekhar Verma, IIIT Allahabad, India
Pabitra Mitra, IIT Kharagpur, India
Joydeep Chandra, IIT Patna, India
Yu-Chen Hu, Providence University, Taiwan
v
vi Conference Committee Members
G. Yvette, Iloilo Science and Technology University, Philippines
Rosslin Robles, University of San Agustin, Philippines
Hamed Daie Kasmaie, Islamic Azad University, Tehran, Iran
Alexey Setekein, Immanuel Kant Baltic Federal University, Russia
Paul S. Pang, Unitec Institute of Technology, New Zealand
Prabhat Mahanti, University of New Brunswick, Canada
Richard G. Bush, Baker College, Michigan, USA
Sayan K. Ray, Manukau Institute of Technology, Manukau, New Zealand
Soocheol Kim, Daegu Catholic University, South Korea
N. Thirupathi Rao, Vignan’s Institute of Information Technology, India
Mohammed Usman, King Khalid University, Abha, Saudi Arabia
Oscar Cordon, Digital University, University of Granada, Spain
Tarek Sobh, University of Bridgeport, Connecticut, USA
Xiao-Zhi Gao, University of Eastern Finland, Finland
Tseren-Onolt Ishdorj, Mongolian University of Science and Technology, Mongolia
Khuder Altangerel, Mongolian University of Science and Technology, Mongolia
Randy Santocildes Tolentino, Hanseo University, South Korea
Diego Galar, Lulea University of Technology, Sweden
Divya Midhunchakkaravarthy, Lincoln University College, Malaysia
C. V. Jawhar, IIIT, Hyderabad, India
Alexander Gelbukh, National Polytechnic Institute, Mexico
Saptarshi Das, Pennsylvania State University, USA
Sourav Sen Gupta, NTU, Singapore
Roumen Koumtchev, Technical University of Sofia, Bulgaria
Christopher Lazarus, Tunku Abdul Rahman University College, Malaysia
Fatiha Merazka, University of Science and Technology Houari Boumediene, Algeria
JinAn Xu, Beijing Jiaotong University, China
Sebastian Ventura Soto, Universidad de Cordoba, Spain
Koneru Satyanarayana, KLEF, Guntur, India
K. Siva Kanchana Latha, KLEF, Guntur, India
Koneru Lakshman Havish, KLEF, Guntur, India
Koneru Raja Hareen, KLEF, Guntur, India
K. S. Jagannatha Rao, KLEF, Guntur, India
G. Pardha Saradhi Varma, KLEF, Guntur, India
N. Venkatram, KLEF, Guntur, India
Megha Bhushan, DIT University, Dehradun, India
Editorial Board
Debnath Bhattacharyya, KL University, Guntur, India
Kingsley A. Ogudo, University of Johannesburg, South Africa
Sanjoy Kumar Saha, Jadavpur University, Kolkata
Conference Committee Members vii
Programme Chairs
Pelin Angin, Middle East Technical University, Turkey
S. Sagar Imambi, K. L. Deemed to be University, Guntur, India
Management Co-chairs
Sonia Djebali, ESILV—Ecole Supérieure d’Ingénieurs Léonard de Vinci, Paris,
France
V. Srikanth, KL University, Guntur, India
Publicity Committee
K. Ravindranath, KL University, Guntur, India
P. Vidya Sagar, KL University, Guntur, India
M. Nageswara Rao, KL University, Guntur, India
Venkata Naresh Mandhala, KL University, Guntur, India
Finance Committee
Venkata Naresh Mandhala, KL University, Guntur, India
Local Arrangements Committee
P. S. V. S. Sridhar, KL University, Guntur, India
K. V. Raju, KL University, Guntur, India
K. Swarna, KL University, Guntur, India
Technical Programme Committee
Sanjoy Kumar Saha, Professor, Jadavpur University, Kolkata
Hans Werner, Associate Professor, University of Munich, Munich, German
Goutam Saha, Scientist, CDAC, Kolkata, India
Samir Kumar Bandyopadhyay, Professor, University of Calcutta, India
viii Conference Committee Members
Ronnie D. Caytiles, Associate Professor, Hannam University, Republic of Korea
Y. Byun, Professor, Jeju National University, Jeju Island, Republic of Korea
Alhad Kuwadekar, Professor, University of South Wales, UK
Debasri Chakraborty, Asst. Professor, BIET, Suri, West Bengal, India
Poulami Das, Assistant Professor, Heritage Institute of Technology, Kolkata, India
Indra Kanta Maitra, Associate Professor, St. Xavier’s University, Kolkata, India
Divya Midhun Chakravarty, Professor, LUC, KL, Malaysia
F. C. Morabito, Professor, Mediterranea University of Reggio Calabria, Reggio
Calabria RC, Italy
Hamed Kasmaei, Islamic Azad University, Tehran, Iran
Nancy A. Bonner, University of Mary Hardin-Baylor, Belton, TX 76513, USA
Alfonsas Misevicius, Professor, Kaunas University of Technology, Lithuania
Ratul Bhattacharjee, AxiomSL, Singapore
Lunjin Lu, Professor, Computer Science and Engineering, Oakland University,
Rochester, MI 48309-4401, USA
Ajay Deshpande, CTO, Rakya Technologies, Pune, India
Debapriya Hazra, Jeju National University, Jeju Island, South Korea
Alexandra Branzan Albu, University of Victoria, Victoria, Canada
G. Yvette, Iloilo Science and Technology University, Philippines
M. H. M. Krishna Prasad, Professor, UCEK, JNTUK, Kakinada, India
N. Thirupathi Rao, Associate Professor, Vignan’s Institute of Information Tech-
nology, Visakhapatnam-530049, India
P. Kishore Kumar, Associate Professor, Vignan’s Institute of Information Tech-
nology, Visakhapatnam-530049, India
Joydeep Chandra, Assistant Professor, IIT Patna, Patna, India
A. Maiti, Assistant Professor, IIT Patna, Patna, India
Jianbang Du, Texas Southern University, USA
Richard G., Bush, Baker College, Michigan, USA
Sourav Sen Gupta, NTU, Singapore
Rosslin J. Robles, Associate Professor, University of San Agustin, Philippines
Jason Levy, University of Hawaii, USA
Daniel Ruiz Fernandez, University of Alicante, Spain
Christo El Morr, York University, Canada
Sayan K. Ray, Manukau Institute of Technology, Manukau, New Zealand
Sanjoy Kumar Saha, Jadavpur University, Kolkata, India
Rinkle Rani, Thapar University, India
G. Neelima, Associate Professor, VIIT, Visakhapatnam, India
Kalpdrum Passi, Laurentian University, Canada
Wafa Shafqat, Jeju National University, South Korea
Alexey Seteikin, Immanuel Kant Baltic Federal University, Russia
Zhang Yu, Harbin University of Commerce, China
Arindam Biswas, Kazi Nazrul University, West Bengal, India
Preface
Knowledge in engineering sciences is about sharing our ideas of research to others. In
engineering, it has many ways to exhibit. In that conference is the best way to propose
your idea of research and its future scope and to add energy to build a strong and
innovative future. So, here we are to give a small support from our side to confer your
ideas by an “International Conference on Smart Technologies in Data Science and
Communication (SMART-DSC 2021)” related to electrical, electronics, information
technology and computer science. It is not confined to a specific topic or region,
and you can exhibit your ideas in similar or mixed or related technologies bloomed
from anywhere around the world because “An idea can change the future and its
implementation can build it”. KLEF Deemed to be University is a great platform to
make your idea(s) penetrated into world. We give as best as we can in every aspect
related. Our environment leads you to a path on your idea, our people will lead your
confidence, and finally, we give our best to make yours. Our intention is to make
intelligence in engineering to fly higher and higher. That is why we are dropping our
completeness into event. You can trust us on your confidentiality. Our review process
is double blinded through Easy Chair.
At last, we pay the highest regard to the Koneru Lakshmaiah Education Foun-
dation, K. L. Deemed to be University from Guntur and Hyderabad, for extending
support for financial management of 5th SMART-DSC 2022.
Guntur, India
Johannesburg, South Africa
Kolkata, India
Debnath Bhattacharyya
Kingsley A. Ogudo
Sanjoy Kumar Saha
ix
Acknowledgements
The editors wish to extend heartfelt acknowledgement to all contributing authors,
esteemed reviewers for their timely response, members of the various organizing
committee and production staff whose diligent work puts a shape to the 5th SMART-
DSC 2022 proceedings. We especially thank our dedicated reviewers for their volun-
teering efforts to check the manuscript thoroughly to maintain the technical quality
and for useful suggestions.
We thank all the following invited speakers who extended their support by sharing
knowledge in their area of expertise.
Prof. Philippe Fournier-Viger, Harbin Institute of Technology (Shenzhen), Shen-
zhen, Guangdong, China.
Prof. Jose L. Seño, Chair, Computer Science Department, College of Information
and Computing Sciences, University of Santo Tomas, Philippines.
Dr. Khuder Altangerel, Head of Computer Science Department, School of
Information, Communication Technology, Mongolian University of Science and
Technology, Ulaanbaatar, Mongolia 14191.
Dr. Shumaila Javaid, Shanghai Research Institute for Intelligent Autonomous
Systems, Tongji University, Shanghai, China.
Dr. Djebali Sonia, ESILV—Ecole Supérieure d’Ingénieurs Léonard de Vinci,
Paris, France.
Dr. Pelin Angin, Middle East Technical University, Ankara, Turkey.
Divya Midhunchakkaravarthy, Lincoln University College, Faculty of Computer
Science and Multimedia, Selangor, Malaysia.
Dr. Snehanshu Pal, National Institute of Technology Rourkela, Rourkela, Odisha,
India.
Debnath Bhattacharyya
Kingsley A. Ogudo
Sanjoy Kumar Saha
xi
Contents
A Graph-Based Model for Discovering Host-Based Hook Attacks ...... 1
P. Pandiaraja, K. Muthumanickam, and R. Palani Kumar
E-Health Care Patient Information Retrieval and Monitoring
System Using SVM ................................................ 15
K. Sumathi and P. Pandiaraja
Number Plate Recognition Using Optical Character Recognition
(OCA) and Connected Component Analysis (CCA) ................... 29
Puppala Ramya, Tummala Haswanth Chowdary,
Pisupati Krishna Teja, and Tadepally Hrushikesh
Cartoonify an Image with OpenCV Using Python .................... 41
Puppala Ramya, Penki Ganesh, Kopanathi Mouli,
and Vutla Naga Sai Akhil
Web Design as an Important Factor in the Success of a Website ........ 51
Puppala Ramya, K. Jai Sai Chaitanya, S. K. Fardeen, and G. Prabhakar
Earlier Selection of Routes for Data Transfer In Both Wired
and Wireless Networks ............................................. 61
S. NagaMallik Raj, S. Neeraja, N. Thirupathi Rao,
and Debnath Bhattacharyya
Identifying River Drainage Characteristics by Deep Neural
Network .......................................................... 71
Vithya Ganesan, Tejaswi Talluru, Manoj Challapalli,
and Chandana Seelam
A Review on Optimal Deep Learning Based Prediction Model
for Multi Disease Prediction ........................................ 81
Aneel Kumar Minda and Vithya Ganesan
xiii
xiv Contents
A Hybrid Multi-user Based Data Replication and Access Control
Mechanism for Cloud Data Security ................................. 91
V. Devi Satya Sri and Srikanth Vemuru
Leveraging the Goldfinger Attack in Blockchain Based
on the Topological Properties ....................................... 101
Arcel Kalenga Muteba and Kingsley A. Ogudo
Bitcoin Transaction Computational Efficiency and Network Node
Power Consumption Prediction Using an Artificial Neural Network .... 109
Arcel Kalenga Muteba, Kingsley A. Ogudo, and Espoir M. M. Bondo
Remote Breast Cancer Patient Monitoring System: An Extensive
Review ........................................................... 117
Sangeeta Parshionikar and Debnath Bhattacharyya
Simplifying the Code Editor Using MEAN Stack Technologies ......... 129
S. NagaMallik Raj, M. Jyothsna, P. Srinu, S. Karthik,
K. Gnana Jeevana, N. Thirupathi Rao, and Debnath Bhattacharyya
Prediction and Identification of Diseases to the Crops Using
Machine Learning ................................................. 139
S. NagaMallik Raj, Pyla Lohit, Doddala Jyo-theendra,
Kannuru Chandana, P. Nikhil, N. Thirupathi Rao,
and Debnath Bhattacharyya
Pulse-Based Smart Electricity Meter Using Raspberry Pi
and MEFN ........................................................ 147
Eswar Abisheak Tadiparthi, Majji Prasanna Kumari,
Basanaboyana Vamsi Sai, Kollana Bharat Kalyan, B. Dinesh Reddy,
N. Thirupathi Rao, and Debnath Bhattacharyya
Brain Tumor Segmentation Using U-Net ............................. 153
Paturi Jyothsna, Mamidi Sai Sri Venkata Spandhana, Rayi Jayasri,
Nirujogi Venkata Sai Sandeep, K. Swathi, N. Marline Joys Kumari,
N. Thirupathi Rao, and Debnath Bhattacharyya
An Empirical Study of CNN-Deep Learning Models for Detection
of Covid-19 Using Chest X-Ray Images .............................. 161
Mohd. Abdul Muqeet, Quazi Mateenuddin Hameeduddin,
B. Mohammed Ismail, Ali Baig Mohammad, Shaik Qadeer,
and M. Muzammil Parvez
Detection of Eye Blink Using SVM Classifier ......................... 171
Varaha Sai Adireddi, Charan Naga Santhu Jagadeesh Boddeda,
Devi Shanthisree Kumpatla, Chris Daniel Mantri, B. Dinesh Reddy,
G. Geetha, N. Thirupathi Rao, and Debnath Bhattacharyya
Contents xv
A Novel Approach for Health Analysis Using Machine Learning
Approaches ....................................................... 179
Debdatta Bhattacharya, N. Thirupathi Rao, K. Asish Vardhan,
and Eali Stephen Neal Joshua
Classification of Healthy and Diseased Lungs by Pneumonia Using
X-Rays and Gene Sequencing With Deep Learning Approaches ........ 189
Debdatta Bhattacharya, K. V. Satyanarayana, N. Thirupathi Rao,
and Eali Stephen Neal Joshua
Breast Cancer Classification Using Improved Fuzzy C-Means
Algorithm ........................................................ 197
N. Thirupathi Rao, K. V. Satyanarayana, M. Satyanarayana,
Eali Stephen Neal Joshua, and Debnath Bhattacharyya
Repercussions of Incorporating Filters in CNN Model to Boost
the Diagnostic Ability of SARS-CoV-2 Virus Using Chest
Computed Tomography Scans ...................................... 205
Dhiren Dommeti, Siva Rama Krishna Nallapati, P. V. V. S. Srinivas,
and Venkata Naresh Mandhala
Software Development Estimation Cost Using ANN ................... 215
Puppala Ramya, M. Sai Mokshith, M. Abdul Rahman, and N. Nithin Sai
A Generic Flow of Cyber-Physical systems—A Comprehensive
Survey ............................................................ 223
Jampani Satish Babu, Gonuguntla Krishna Mohan, and N. Praveena
Mental Disorder Detection in Social Networks Using SVM
Classification: An Improvised Approach ............................. 241
B. Dinesh Reddy, Eali Stephen Neal Joshua, N. Thirupathi Rao,
and Debnath Bhattacharyya
An Enhanced K-Means Clustering Algorithm to Improve
the Accuracy of Clustering Using Centroid Identification Based
on Compactness Factor ............................................ 251
Eali Stephen Neal Joshua, K. Asish Vardhan, N. Thirupathi Rao,
and Debnath Bhattacharyya
Prediction of Chronic Kidney Disease with Various Machine
Learning Techniques: A Comparative Study ......................... 257
K. Swathi and G. Vamsi Krishna
Blockchain and Its Idiosyncratic Effects on Energy Consumption
and Conservation .................................................. 263
K. Mrudula Devi, D. Surya Sai, N. Thirupathi Rao, K. Swathi,
and Swathi Voddi
xvi Contents
Smart Hydroponics System for Soilless Farming Based on Internet
of Things ......................................................... 271
G. V. Danush Ranganath, R. Hari Sri Rameasvar, and A. Karthikeyan
Solution Approach for Detection of Stock Price Manipulation
by Market Operators .............................................. 281
Yogesh Kakde, Ganesh Chavan, Basant Sah, and Apoorva Sen
Cancer Cell Detection and Classification from Digital Whole Slide
Image ............................................................ 289
Anil B. Gavade, Rajendra B. Nerli, Shridhar Ghagane,
Priyanka A. Gavade, and Venkata Siva Prasad Bhagavatula
Author Index ...................................................... 301
Editors and Contributors
About the Editors
Prof. (Dr.) Kingsley A. Ogudo, Ph.D. received the Master in Electrical Elec-
tronics/telecommunication engineering and Doctoral Degree in electrical and elec-
tronics engineering technology from the Tshwane University of Technology (TUT),
South Africa, in 2010 and 2016 respectively. He received his Ph.D. in Electronics
and Optoelectronics systems from the University of Paris Est, France in July year
2018. His research interest includes electronic, optoelectronic devices, Power elec-
tronics; System Integration of Devices based on Renewable Energy management
Sources, Telecommunication engineering high-frequency electronics, AI, IoT and
Data Analytics, physics and applied mathematics. He is a Professional Engineer
Technologist certified by ECSA and he is a member of the IEEE Society. He is a
Fellow at SAIEE, and Secretary General for SAIEE Entrepreneur and innovation
chapter. He has lectured in three different university (Tshwane University of Tech-
nology, UNISA and UJ) for the past 11 years, and has thought different electrical
and electronics engineering modules (Subjects) to both undergraduates and post-
graduates students. He has published over 65 international ISI Journal articles and
international conference papers. He is currently an Associate Professor/Researcher
at the Department of Electrical and Electronics Engineering Technology, University
of Johannesburg (UJ), South Africa.
Dr. Sanjoy Kumar Saha currently associated as Professor with the Department
of Computer Science and Engineering, Jadavpur University, Kolkata, India. He
did is BE and ME from Jadavpur University and completed his Ph.D. from IIEST
Shibpur, West Bengal, India. His Research interests include Image, Video and Audio
Data Processing, Physiological Sensor Signal Processing and Data Analytics. He
published more than hundred articles in various International Journals and Confer-
ences of repute. He has guided eleven Ph.D. Students. He holds four US patents. Dr.
Saha is a member of IEEE Computer Society, Indian Unit for Pattern Recognition
xvii
xviii Editors and Contributors
and Artificial Intelligence, ACM. He has served TCS innovation Lab, Kolkata, India
as advisor for the signal processing group.
Prof. (Dr.) Debnath Bhattacharyya is associated as a Professor with Computer
Science and Engineering Department, Koneru Lakshmaiah Education Foundation
(known as K. L. Deemed to be University), Guntur, Andhra Pradesh, India. Dr. Bhat-
tacharyya is presently an Invited/Visiting International Professor, Lincoln University
College, KL, Malaysia and University of Johannesburg, South Africa. Dr. Bhat-
tacharyya received his Ph.D. (Tech., Computer Science and Engineering) from the
University of Calcutta, Kolkata, India. Dr. Bhattacharyya is the Senior Member of
IEEE, Member of ACM, and Life Member of CSI, India. He is the Editor of Many
International Journals (indexed by Scopus, SCI, and Web of Science). He Published
234 Scopus Indexed Papers, and 145 Web of Science Papers. His Research inter-
ests include Security Engineering, Pattern Recognition, Biometric Authentication,
Multimodal Biometric Authentication, Data Mining and Image Processing.
Contributors
Abdul Rahman M. Department of Computer Science and Engineering, Koneru
Lakshmaiah Education Foundation, Vaddeswaram, India
Adireddi Varaha Sai Department of Computer Science and Engineering, Vignan’s
Institute of Information Technology (A), Visakhapatnam, Andhra Pradesh, India
Akhil Vutla Naga Sai Department of Computer Science and Engineering, Koneru
Lakshmaiah Education Foundation, Guntur, India
Asish Vardhan K. Department of Computer Science and Engineering, Bullayya
College of Engineering for Women, Visakhapatnam, Andhra Pradesh, India
Babu Jampani Satish Department of Computer Science and Engineering, Koneru
Lakshmaiah Education Foundation, Vaddeswaram, Guntur, A.P, India
Bhagavatula Venkata Siva Prasad Medtronic, Hyderabad, India
Bhattacharya Debdatta Department of Computer Science and Engineering,
Koneru Laksmaiah Education Foundation, Vaddeswaram, Guntur, Andhra Pradesh,
India
Bhattacharyya Debnath Department of Computer Science and Engineering,
Koneru Lakshmaiah Education Foundation, Vaddeswaram, Guntur, Andhra Pradesh,
India
Boddeda Charan Naga Santhu Jagadeesh Department of Computer Science and
Engineering, Vignan’s Institute of Information Technology (A), Visakhapatnam,
Andhra Pradesh, India
Editors and Contributors xix
Bondo Espoir M. M. Engineering Research and Development BOND’AF, Paris,
France
Challapalli Manoj CSE, Koneru Lakshmaiah Education Foundation, Guntur,
Andhra Pradesh, India
Chandana Kannuru Department of CSE, Vignan’s Institute of Information Tech-
nology (A), Duvvada, Visakhapatnam, India
Chavan Ganesh KL University, Guntur, AP, India
Chowdary Tummala Haswanth Department of Computer Science and Engi-
neering, Koneru Lakshmaiah Education Foundation, Guntur, India
Danush Ranganath G. V. School of Electrical Engineering, Vellore Institute of
Technology, Vellore, India
Devi Satya Sri V. Department of Computer Science and Engineering, Koneru
Lakshmaiah Education Foundation, Vaddeswaram, Guntur District, A.P, India
Devi K. Mrudula Department of Mathematics, Vignan’s Institute of Information
Technology (A), Visakhapatnam, AP, India
Dinesh Reddy B. Department of Computer Science and Engineering, Vignan’s
Institute of Information Technology, Visakhapatnam, Andhra-Pradesh, India
Dommeti Dhiren Department of Computer Science Engineering, Koneru Laksh-
maiah Education Foundation, Vaddeswaram, Andhra Pradesh, India
Fardeen S. K. Department of Computer Science and Engineering, Koneru Laksh-
maiah Education Foundation, Guntur, India
Ganesan Vithya CSE, Koneru Lakshmaiah Education Foundation, Guntur, Andhra
Pradesh, India
Ganesh Penki Department of Computer Science and Engineering, Koneru Laksh-
maiah Education Foundation, Guntur, India
Gavade Anil B. Department of E&C, KLS Gogte Institute of Technology, Belagavi,
Karnataka, India
Gavade Priyanka A. Department of Computer Science and Engineering, KLE
Tech University Dr. M. S. Sheshgiri College of Engineering and Technology,
Belagavi, Karnataka, India
Geetha G. Department of Information Technology, VR Siddhartha Engineering
College, Kanuru, Vijayawada, Andhra Pradesh, India
Ghagane Shridhar Department of Biotechnology, KAHER’s Dr. Prabhakar Kore
Basic Science Research Center, V. K. Institute of Dental Sciences Campus, Belagavi,
Karnataka, India
xx Editors and Contributors
Hameeduddin Quazi Mateenuddin Faculty of Electronics and Communication
Engineering, Indian Naval Academy, Ezhimala, Kerala, India
Hari Sri Rameasvar R. School of Electrical Engineering, Vellore Institute of
Technology, Vellore, India
Hrushikesh Tadepally Department of Computer Science and Engineering, Koneru
Lakshmaiah Education Foundation, Guntur, India
Jai Sai Chaitanya K. Department of Computer Science and Engineering, Koneru
Lakshmaiah Education Foundation, Guntur, India
Jayasri Rayi Department of Computer Science & Engineering, Vignan’s Institute
of Information Technology (A), Visakhapatnam, AP, India
Jeevana K. Gnana Department of CSE, Vignan’s Institute of Information Tech-
nology (A), Duvvada, Visakhapatnam, India
Joshua Eali Stephen Neal Department of Computer Science and Engineering,
Vignan’s Institute of Information Technology, Visakhapatnam, Andhra Pradesh,
India
Jyo-theendra Doddala Department of CSE, Vignan’s Institute of Information
Technology (A), Duvvada, Visakhapatnam, India
Jyothsna M. Department of CSE, Vignan’s Institute of Information Technology
(A), Duvvada, Visakhapatnam, India
Jyothsna Paturi Department of Computer Science & Engineering, Vignan’s Insti-
tute of Information Technology (A), Visakhapatnam, AP, India
Kakde Yogesh KL University, Guntur, AP, India
Kalyan Kollana Bharat Department of Computer Science and Engineering,
Vignan’s Institute of Information Technology, Visakhapatnam, Andhra-Pradesh,
India
Karthik S. Department of CSE, Vignan’s Institute of Information Technology (A),
Duvvada, Visakhapatnam, India
Karthikeyan A. School of Electrical Engineering, Vellore Institute of Technology,
Vellore, India
Kumari Majji Prasanna Department of Computer Science and Engineering,
Vignan’s Institute of Information Technology, Visakhapatnam, Andhra-Pradesh,
India
Kumpatla Devi Shanthisree Department of Computer Science and Engineering,
Vignan’s Institute of Information Technology (A), Visakhapatnam, Andhra Pradesh,
India
Lohit Pyla Department of CSE, Vignan’s Institute of Information Technology (A),
Duvvada, Visakhapatnam, India
Editors and Contributors xxi
Mandhala Venkata Naresh Department of Computer Science Engineering,
Koneru Lakshmaiah Education Foundation, Vaddeswaram, Andhra Pradesh, India
Mantri Chris Daniel Department of Computer Science and Engineering, Vignan’s
Institute of Information Technology (A), Visakhapatnam, Andhra Pradesh, India
Marline Joys Kumari N. Department of Computer Science & Engineering, Anil
Neerukonda Institute of Technology and Sciences, Visakhapatnam, AP, India
Minda Aneel Kumar International SOS, Dubai, United Arab Emirates
Mohammad Ali Baig School of Electronics and Communication Engineering,
REVA University, Bengaluru, India
Mohammed Ismail B. Department of Artificial Intelligence & Machine Learning,
P.A. College of Engineering, Mangalore, Karnataka, India
Mohan Gonuguntla Krishna Department of Computer Science and Engineering,
Koneru Lakshmaiah Education Foundation, Vaddeswaram, Guntur, A.P, India
Mouli Kopanathi Department of Computer Science and Engineering, Koneru
Lakshmaiah Education Foundation, Guntur, India
Muqeet Mohd. Abdul Electrical Engineering Department, Muffakham Jah
College of Engineering and Technology, Hyderabad, India
Muteba Arcel Kalenga Department of Electrical and Electronics Engineering
Technology, University of Johannesburg, Johannesburg, South Africa
Muthumanickam K. Department of Information Technology, Kongunadu College
of Engineering and Technology (Autonomous), Thottiyam, Tiruchirappalli, India
Muzammil Parvez M. Electronics and Communication Engineering Department,
KLEF, Deemed to Be University, Vaddeswaram, A.P, India
NagaMallik Raj S. Department of Computer Science & Engineering, Vignan’s
Institute of Information Technology (A), Visakhapatnam, Andhra Pradesh, India
Nallapati Siva Rama Krishna Department of Computer Science Engineering,
Koneru Lakshmaiah Education Foundation, Vaddeswaram, Andhra Pradesh, India
Neal Joshua Eali Stephen Department of Computer Science and Engineering,
Vignan’s Institute of Information Technology, Visakhapatnam, Andhra Pradesh,
India
Neeraja S. Department of Computer Science &, Software Engineering Lendi
Institute of Engineering and Technology, Vizianagaram, Andhra Pradesh, India
Nerli Rajendra B. Department of Urology, JN Medical College, KLE Academy of
Higher Education and Research (Deemed-to-Be-University), Belagavi, Karnataka,
India
xxii Editors and Contributors
Nikhil P. Department of CSE, Vignan’s Institute of Information Technology (A),
Duvvada, Visakhapatnam, India
Nithin Sai N. Department of Computer Science and Engineering, Koneru Laksh-
maiah Education Foundation, Vaddeswaram, India
Ogudo Kingsley A. Department of Electrical and Electronics Engineering Tech-
nology, University of Johannesburg, Johannesburg, South Africa
Palani Kumar R. Department of Information Technology, Kongunadu College of
Engineering and Technology (Autonomous), Thottiyam, Tiruchirappalli, India
Pandiaraja P. Department of Computer Science and Engineering, M.Kumarasamy
College of Engineering, Thalavapalayam, Karur, TamilNadu, India
Parshionikar Sangeeta Department of Computer Science and Engineering,
Koneru Lakshmaiah Education Foundation, KLEF, Guntur, Andhra Pradesh, India
Prabhakar G. Department of Computer Science and Engineering, Koneru Laksh-
maiah Education Foundation, Guntur, India
Praveena N. VR Siddharth Engineering College, Kanuru, Vijayawada, AP, India
Qadeer Shaik Electrical Engineering Department, Muffakham Jah College of
Engineering and Technology, Hyderabad, India
Ramya Puppala Department of Computer Science and Engineering, Koneru
Lakshmaiah Education Foundation, Vaddeswaram, Guntur, India
Rao N. Thirupathi Department of Computer Science and Engineering, Vignan’s
Institute of Information Technology (A), Visakhapatnam, AP, India
Reddy B. Dinesh Department of Computer Science and Engineering, Vignan’s
Institute of Information Technology (A), Visakhapatnam, Andhra Pradesh, India
Sah Basant KL University, Guntur, AP, India
Sai Mokshith M. Department of Computer Science and Engineering, Koneru
Lakshmaiah Education Foundation, Vaddeswaram, India
Sai Basanaboyana Vamsi Department of Computer Science and Engineering,
Vignan’s Institute of Information Technology, Visakhapatnam, Andhra-Pradesh,
India
Sai D. Surya Department of Computer Science and Engineering, Vignan’s Institute
of Information Technology (A), Visakhapatnam, AP, India
Sandeep Nirujogi Venkata Sai Department of Computer Science & Engineering,
Vignan’s Institute of Information Technology (A), Visakhapatnam, AP, India
Satyanarayana K. V. Department of Computer Science and Engineering, Raghu
Engineering College, Visakhapatnam, Andhra Pradesh, India
Editors and Contributors xxiii
Satyanarayana M. Department of Computer Science and Engineering, Raghu
Engineering College, Visakhapatnam, Andhra Pradesh, India
Seelam Chandana CSE, Koneru Lakshmaiah Education Foundation, Guntur,
Andhra Pradesh, India
Sen Apoorva Medi-Caps University, Indore, MP, India
Spandhana Mamidi Sai Sri Venkata Department of Computer Science & Engi-
neering, Vignan’s Institute of Information Technology (A), Visakhapatnam, AP,
India
SrinivasP.V.V.S. Department of Computer Science Engineering, Koneru Laksh-
maiah Education Foundation, Vaddeswaram, Andhra Pradesh, India
Srinu P. Department of CSE, Vignan’s Institute of Information Technology (A),
Duvvada, Visakhapatnam, India
Sumathi K. Department of Computer Science and Engineering, KSR Institute for
Engineering and Technology, Thiruchencode, TamilNadu, India
Swathi K. Department of Computer Science and Engineering, Vignan’s Institute of
Information Technology (A), Visakhapatnam, AP, India
Tadiparthi Eswar Abisheak Department of Computer Science and Engineering,
Vignan’s Institute of Information Technology, Visakhapatnam, Andhra-Pradesh,
India
Tall u r u Te j a s wi CSE, Koneru Lakshmaiah Education Foundation, Guntur, Andhra
Pradesh, India
Teja Pisupati Krishna Department of Computer Science and Engineering, Koneru
Lakshmaiah Education Foundation, Guntur, India
Vamsi Krishna G. Department of Computer Science and Engineering, Dr. Lanka-
palli Bullayya College of Engineering, Visakhapatnam, AP, India
Vardhan K. Asish Department of Computer Science and Engineering, Bullayya
College of Engineering for Women, Visakhapatnam, AP, India
Vemuru Srikanth Department of Computer Science and Engineering, Koneru
Lakshmaiah Education Foundation, Vaddeswaram, Guntur District, A.P, India
Voddi Swathi Department of Computer Science and Engineering, Prasad V. Potluri
Siddhartha Institute of Technology, Vijayawada, Andhra Pradesh, India
A Graph-Based Model for Discovering
Host-Based Hook Attacks
P. Pandiaraja, K. Muthumanickam, and R. Palani Kumar
Abstract Though computer malicious software can be referred with different
names such as virus, worm, Trojan, spam, and botnet, their ultimate goal is to
causing damage to the end-computer or end-user. The progression in computer tech-
nology allows a malware writer to integrate obfuscation technique to evade detec-
tion specifically API hooking in Windows. Unfortunately, signature-based detection
approach such as anti-virus software at the end-computer is not effective against
system call reordering. To overcome this shortcoming, many different behavior-based
approaches have been offered. However, these approaches bear limitations such as
false positive, detecting zero-day attacks, and improving detection accuracy rate from
past experience. In this article, an application programming interface (API)-based
call graph model is put forward which captures API system call during malicious
rootkit execution in Windows platform. As graph model can be effectively applied to
replica complicated relation between entities, we opt it to visualize malicious rootkit
behavior activities by monitoring system API calls. This will help the defender to
optimally find malicious system calls from benign calls. Our simulated experiment
analysis proves that our method achieves higher detection rate and accuracy with
less false positive compared to existing techniques.
Keywords API hook ·Graph ·Malware attack ·Rootkit
P. Pandiaraja (B)
Department of Computer Science and Engineering, M.Kumarasamy College of Engineering,
Karur, Tamil Nadu 639113, India
e-mail: sppandiaraja@gmail.com
K. Muthumanickam · R. Palani Kumar
Department of Information Technology, Kongunadu College of Engineering and Technology
(Autonomous), Thottiyam, Tiruchirappalli 621215, India
e-mail: muthumanickam@kongunadu.ac.in
R. Palani Kumar
e-mail: palanikumar@kongunadu.ac.in
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_1
1
2 P. Pandiaraja et al.
1 Introduction
Today hacktivists and cyber-criminals can able to write malwares with advanced
evading techniques and continuing to evolve different techniques with the intent of
assaulting end-user’s privacy. A new type of malware will be launched every day by
modifying its predecessor. The AV-Test report [1] detected more than 500 million
malware s amples in 2018. Analyzing such a huge malware sample manually is a
tedious process, so we need to have an automated malware analysis technique to
craft virus definition. Today, malware developers often integrate rootkit technique
which mainly uses an API hook technique into malware software to avoid detection.
A malware detector is a software program that can be operated locally on the victim
computer discover and locate a malware. Usually, there are two different kinds of
inputs given to a malware detector, namely unique signature of the malware or moni-
toring its behavior. After gaining these two inputs, it is easy for a malware detector
to identify malicious programs.
Nowadays, a huge number of malicious samples are submitted frequently to secu-
rity companies for analyzing whether each sample is malware or legitimate. In order
to expose hijacked API calls, we need to have a behavioral monitoring system which
categorizes malicious activities and legitimate activities. Existing tool such as [2,
3] are useful for generating reports on unknown executables which affect Windows
API calls. However, the generated reports manually need to be clustered based on
similar behavior or based on malicious activity. Analyzing a huge amount of malware
infected information to recognize its intended attack is typically a difficult issue. Few
present works have existed that rely on signature relied approach.
Activities of a malware can be detected by collecting anomalous network traffic
for example, and botnet attack can be identified by collectively monitoring a network
of computers and then look for computers that exhibit similar communication pattern.
Though network-based analysis approach is useful, they suffer from several limita-
tions. First, a malware packet may imitate as a legitimate packet to avert detection.
Secondly, if the payload of a malware is encrypted, then collecting and analyzing
network traffic cannot reveal its presence. Thirdly, network-based approach fails to
sense malicious activities when they cannot do communication with remote attacker.
In addition to signature-based approach, another fitting place to supervise and inves-
tigate malware behavior is at the end-host. We can detect a malicious code attack
even before it gets executed in the victim computer. However, current host-based
malicious code detection t echniques do not use effective models. As a result, these
models cannot capture central or essential properties of a malicious executable. An
API call graph (ACG) is a candidate solution which is a suitable data illustration of
the data and control flood of software programs. Additionally, it offers information
about local data usage of a procedure and global data that can be exchanged between
different procedures. Call graph acts as a suitable tool either to study the behavior of a
program or for tracking the flow values between different components of a program.
ACG can also be used to recognize programs that are never invoked. In this paper,
we present an ACG framework for detecting malicious software that uses API hook
A Graph-Based Model for Discovering Host-Based Hook Attacks 3
attacks based on the synthesis of static and dynamic analysis technique. The theme
of this paper is discussed as follows:
Static and dynamic analysis methods are used for the identification and extraction
of API invocation calls and its associated parameters of an executable.
The API system call-dependent graph algorithm is devised to generate graphs
from the extracted information.
Finally, ACG algorithm is implemented to compare all data-dependent graphs
which can identify whether an API call made by the executable is either legitimate
or malicious hook attack.
The arrangement of this article is structured as follows. Section 2 presents the
existing techniques to detect malware attacks using a graph model, and Sect. 3
explains about the proposed system to optimally detect malware attacks. Experi-
mental environment and the evaluation results are analyzed in Sect. 4. Section 5 lists
conclusion.
2 Related Works
Graphs can be used to reflect the execution flow of an executable file through nodes
(vertices) and edges (links) that, respectively, denote API function calls and relation-
ship between API function calls. Almost all recent malwares are being developed
from its predecessor by incorporating new features. The operations to be invoked by
a system call can be traced and modeled as a digraph that is composed of nodes and
edges, where each node signifies a function call and each edge denotes calls between
functions. Such a graph is referred as a call graph.
Malicious rootkit that uses API hook technique continues to be an advancing
hazard to current computing technology. With the ever-growing explosion of these
kinds of threats, it is required to build up a new method to combat them. Though
many antivirus programs are available to classify files as being either malware or
benign, they suffer from two limitations. First, they rely on signature-based approach
which cannot identify unknown malware signature. Secondly, antivirus programs
cannot deal with malware that uses API hook technique. There are many graph-
based approaches that have been proposed in the past to dynamically analyze malware
attacks. The n-gram approach was one of the first methods to spot malware activities
especially identifying polymorphic and obfuscated viruses [6–8]. The uniqueness of
our work is to identify API hook attacks in a novel way which utilizes a graph-based
approach.
A graph is an attractive tool for analyzing malware hits efficiently [8, 10]. In order
to investigate malware-based attacks in the Internet, Red team manually generating
graphs. But their work has either false positive or difficult for a malicious malware
that implements API hook technique. So, researchers are using different technique
code graph or call graph to build and analyze malware attacks [4]. Guo et al. [9]
proposed a binary translation approach to analyze and detect malware execution.
The authors generated control flow graph based on malware’s behavior, and then
4 P. Pandiaraja et al.
another API subgraph was generated to compare its activities. The authors in paper
[5] presented a graph-based malware inference model that relied on system call
information which can be invoked at the time of execution in a victim computer.
This method offered improved detection rate and avoids scalability issue. Many
works published in the past stress the importance of concern machine learning and
statistical methods to discover the presence of a malware. Nath and Mehtre [11]
proposes a mixture of different data samples which can be created from malicious
malware trials for detecting malware, like n-grams, instructions, and unique byte
string.
Bio-sequence-based comparison methods also exist [12] for evaluating genetic
trails which relies on genetic chain, to detect legitimate executables. Cuckoo sandbox
[13] is a most popular malware analysis tool. This open-source tool can be used
to automatically analyze many different files like emails, executables, etc., and
infer informative data. These data summarize the flow direction of execution of
the malware and collect information about API function calls, registry file, and flow
of network traffics. Pirscoveanu et al. [14] utilized cuckoo to achieve improved
classification rate. Elhadi et al. [15] developed an API call graph model using depen-
dency relationship and profile of function calls to discover malicious operations. This
model uses past history of known discovered malware samples to identify unknown
malware attacks. However, polymorphic packed malware would make detecting
zero-day attacks very multifarious. Mehra et al. [16] proposed a combination of
control flow graph (CFG) API call graph (ACG) and histogram technique to classify
a system as wither benign or malicious. This method uses two different algorithms:
one for removing unwanted data and to manipulate a CFG and another algorithm for
generating ACG and its features.
The modified longest common subsequence algorithm (m-LCSA) [16] is utilized
to find out the similarity linking two strings using the longest subsequences that
are common to all input strings and determine best subgraph. Khodamoradi et al.
[17] applied the decision tree method to infer statistical information on opcode from
disassembled code and then build threshold values. The opcode statistic extractor
tool is used for examining disassembled code to calculate frequency value of the
opcode which was then considered to check whether malicious code is present. Mosli
et al. [18] proposed a machine learning-based malware detection approach using
support vector machine. This method extracts different features like API function
calls, registry access, and import/export library functions from malware accessed
memory area.
An existing method [19] deployed hybrid solutions that apply various stemming
techniques and algorithms to optimize detection accuracy. Kane et al. [20] proposed
an optimized opcode method for discovering obfuscated malicious executables. In
this work, first, support vector machine technique was applied to categorize different
type of files. Then, a histogram-based opcode density extraction procedure was exer-
cised to create opcode set during application execution. Salehi et al. [21] presented a
study on generating important features of argument return values about API function
call lists. The experimental results indicate that this research work obtained detection
rate of 99.9% with negligible false positives. Techniques for comparing nodes and
A Graph-Based Model for Discovering Host-Based Hook Attacks 5
structures of two different call graphs and their similarity level will be exercised to
detect malware in this paper. We anticipate system call traces of a function call to be
very similar with similar structures [23]. In addition, unrelated system calls should
invoke some API function calls with dissimilar structures. Few existing research
papers [24] also impose authentication system during validation of system calls of
different applications.
3 Proposed Graph-Based Model
We assume that most malicious malwares are developed by inheriting characteristics
from its previous version. For example, the various versions of TDSS rootkit are as
follows: TDL1 was implemented to load and run at the time of booting the operating
system which was designed with the intention of infecting drivers. TDL2 appears to
be same as TDL1. However, it includes different names with random string and also
imports new technique to avoid detection and removal. In order to obtain control
over the victim computer, TDL3 patches the disk controller driver. Some features of
TDL2 were updated to make detection and removal more difficult. The aim of TDL4
variant is the same as that of TDL3. However, patching Master Boot Record is done
which makes infection of 64-bit systems also possible.
A directed graph G is a call graph, (V, E) in which V signifies a set of nodes
that represent a function of the executable program and E is a set of ordered pairs of
elements, E_VxV [14]. A directed edge (u, v)in E represents a function call of the
program, u > v. The proposed idea attempts to optimize the accuracy of malicious
code API hook attack detection using API call graph. An API call graph is constructed
using data-dependent plot in which each node represents an API call and each edge
denotes the dependencies between two calls.
The modified LCS graph matching algorithm is applied to identify common
subgraphs and their similarity. The overall picture of our system is given in Fig. 1,
which includes two important stages which are referred to as preprocessing stage
and post-processing stage.
3.1 Preliminary Processing Stage
A function can be invoked or called itself to accomplish certain task. An API-based
call graph (ACG) is generated to show the relationship between callees and callers.
An ACG acts as a vital source to extract important features. Important function types
that can be exercised to generate necessary resource of an ACG can be classified as
follows.
nlFun: the API functions which do not reside in the system’s dynamic link library
(dll) and can automatically generate function names.
6 P. Pandiaraja et al.
Fig. 1 Structure of the proposed system
lFun: the API functions which resides in the system’s dll file.
imFun: the API functions which are imported from the system’s dll file.
xFun: the API functions which are not identified as library API functions but use
jump instruction to execute a detoured code indirectly.
There are few malicious hooks which do not follow any of the four aforementioned
API function calls, and discovering such hooks is out of the scope of this article. All
the nodes and edges of an ACG are extracted from the aforementioned API functions.
An edge with some cost for a pair of nodes can be produced using two API-based
calls associated with it. To construct an ACG, all the functions associated with an
executable will be identified by referring the system tables, namely Import Address
Table and Export Address Table. Then each function is verified to determine whether
it is a system call or not. If it is a system call, its corresponding function name and
its parameters are used to construct a call graph. An ACG is generated in which
each node contains the function name and an edge is established using its parameters
list. If two parameters are available in list that are redundant, then they reflect the
dependence between the recent and preceding API-based call.
3.2 Malware Detection Stage
Today, a malware writer can develop a malware by updating new features and tech-
niques with its predecessor rather than coding from scratch. This information helps
us to reduce the complexity of considering all kinds of addiction while inquiring the
A Graph-Based Model for Discovering Host-Based Hook Attacks 7
Table 1 Important features mined from ACG
Feature Comment
Node API function to be invoked through system call
Edge Relationship between two API function calls
Start node Start node in the ACG
Isolated node Function which does not call any other function
Subgraph An undirected subgraph of the ACG
Type of a node A node belongs to any of the function type (nlFun, lFun, imFun, and xFun)
data graph (DG). The objective of the malware detection stage is to generate a subset
of the DG by referring query graph (QG) and extraction of few important features
of such graphs. The important predefined features that are used to detect a malware
sample are given in Table 1.
Definition 1 A subgraph Gx = (Vx, Ex) contains both start node and the last node
recently visited and the edge between these two nodes. A subgraph does not contain
a new subgraph Gy = (Vy, Ey) but Vy ⊆ Vx, where Vx represents the collection of
nodes and Ex refers to the collection of edges.
Definition 2 A best subgraph includes nodes of all the recently generated subgraphs.
The central idea of graph similarity is to generate a subgraph of DG by best matching
the QG. To apply m-LCSA, the data-dependent ACG is required to be transformed
into sequence of a string. The desired algorithm then maps a path of QG against
the path in the DG using m-LCSA. The pseudocode of the m-LCSA is given in
Algorithm 1.
Algorithm 1. Algorithm for matching call graph(s)
1. Input: Query Graph (QG) and Data Graph (DG)
2. Output: Similarity Matching
3. procedure SIMMATCHING(QG;DG)
4. rval ← 0
5. Extract paths of P1 and P2
6. Iu ← list of items rated by Ui
7. if (Paths of P1 and P2 has same label in every edge) then
8. for (every path find similarity using LCSA) do
9. rval ← rval + LCSA(P1; P2)
10. node ← function_name
11. end for
12. end if
13. p ← paths_in_QG
14. r (QG;DG) ← r|p|
15. if (rval(P1) == rval(P2) then
16. ’Malicious API call’
17. end if
18. return rval
19. end procedure
8 P. Pandiaraja et al.
A malware detection scheme is proposed to discover unknown malicious executa-
bles using two stage procedures. First, API function calls to be invoked are modeled
as an ACG. Then, few important features are extracted from t he ACG which can be
used for training the proposed scheme. Finally, the presence of a malware sample is
discovered using features extracted from the ACG.
4 Experimental Results and Discussion
We focused on detecting malwares t hat execute PE executables on Windows platform.
As there is no standard yardstick exists for comparing two graphs to detect a malware
attack, many researchers are using their own malware datasets against various assess-
ment techniques. We have collected a malware samples dataset that contains 250
worms, 250 viruses, 250 Trojans and 250 benign legitimate programs that uses an
API hook technique. Benign programs have been collected from a computer that runs
a fresh copy of Windows 7 and Windows XP. We have run each malware sample
in an isolated environment to identify and extract AP calls and its parameters using
API monitoring tool. The API calls of an executable are identified by analyzing
binary files statistically using tool like IDA Pro [18] or by executing the binary files
dynamically in an isolated environment using a tool like API monitor [19]. Though
API-based calls can be analyzed through dynamic investigation, the malicious binary
must be executed several times to spot various execution flows.
To dynamically analyze a malicious executable files, the following three opera-
tions are performed. First, the obfuscation cover is removed. Secondly, unpacking
and decryption are performed over the executable. Finally, functions are extracted
which are later assigned with a unique symbolic name. Using this extracted infor-
mation, a graph is generated for each API call. We utilize different techniques like
random forest and data mining classification techniques to produce appropriate clas-
sifiers. In order to verify the usefulness of our method in detecting the presence of a
malware, different malware samples with cross-validation method are applied. The
test dataset is partitioned into ten different sets—a set on the average consists of 75
malware samples and 25 benign programs. Then the proposed framework has trained
on nine sets, and the last set is taken for testing it.
In order to utilize call graphs to exactly locate API hook attacks, it is necessary
to compare a call graph that reflects the API hook behavior against those that reflect
benign behavior. To compare two call graphs, we used a graph matching Algorithm
1 to determine its similarity by matching data graphs with query graphs. When two
graphs have the same number of nodes, then it is said to be exact matching. All experi-
ments are tested on machine runs Windows 7 operating system. For every system call,
its equivalent DG is generated. Then it is compared with QG. By analyzing numerous
root malware attacks, we set a similarity threshold value of 95% to determine whether
a generated graph impersonates malicious operation or not. Suppose the determined
similarity cost of a malware surpasses the predefined threshold similarity value, then
it is suspected as a malicious malware that uses API hook attack.
A Graph-Based Model for Discovering Host-Based Hook Attacks 9
4.1 Discussion
Table 2 lists the overall detection accuracy rate of different graph-based approaches
considered for malware detection. The random forest graph-based approach attained
detection accuracy rate (worm) of 97.5% which is only less than 0.4% compared to
detection rate of Trojan. Although worms and Trojans used different kinds attacking
strategies, the detection accuracy rate looks approximate. We came to the same
conclusion from the outcome obtained from the next classifier, data mining. However,
the proposed method attained approximately 99% of detection accuracy and outper-
forms other methods. All the methods listed in Table 2 also obtained nearly the same
detection accuracy rate when different dataset has been used, and this can confirm
the consistency of the proposed model.
The accuracy of our method is appraised by using parameters such as false posi-
tive (FP) that occurs when the test spots the legitimate programs to be malicious,
detection rate (DR), and accuracy rate (AR). The percentage of programs classified
as malicious is measured as false positive rate (FPR) that is determined using the
following formula. FPR = FP/(FP + TN). Figure 2 shows the ROC curves of all
detection of all techniques that have taken for analysis and comparison, and Table 3
depicts the AUC values of each technique.
The small twisted in curve of data mining reveals that data mining-based malware
detection suffers from more false positives. As the AUC values of both random forest
and data mining almost same, their ROC curve almost overlaps. The simulation
results of the proposed method achieve better AUC value, i.e., 99% in all cases of
malware samples than the rest of two techniques with minimal false positives. The
same training and testing datasets are employed in bigrams and graph edit distance-
based approach [22], and its comparison with our approach is presented in Table
2. In order to test the robustness of the proposed scheme, a small dataset (1%) has
been randomly chosen for training purpose which will discover the remaining 99%.
Figure 3 demonstrates the performance of proposed malware detection approach, and
its detection accuracy is compared with the bigrams and graph edit distance-based
approach.
It can be inferred that the detection accuracy of the proposed method achieves
98% when the size of the dataset is 9% of the entire dataset, and 99% can be reached
constantly when the dataset is increased from 10%. The malware detection of the
bigrams and GED-based method has achieved below 95% when the dataset is 1%
Table 2 Detection accuracy
rate of different graph-based
approaches
Approach Detection accuracy (%)
Wor m Virus Trojan
Random forest 97.1 97.3 97.5
Data mining 96.1 96.6 95.5
Proposed approach 98.9 98.7 98.8
Graph edit distance (GED) 97.6 96.7 96.2
10 P. Pandiaraja et al.
Fig. 2 ROC curves of random forest, data mining, and proposed test results
Table. 3 AUC values of
random forest, data mining,
and proposed test results
Approach Wor m Virus Trojan
Random forest 0.984 0.985 0.987
Data mining 0.987 0.981 0.982
Proposed approach 0.992 0.993 0.994
and attained overall malware detection rate 97%. There are two reasons for variation
in detection accuracy. First, different malware dataset is used for training and testing
purpose and second, the bigrams and GED-based method for dependence on the
features of known malware samples.
A Graph-Based Model for Discovering Host-Based Hook Attacks 11
Fig. 3 Performance comparison of the proposed method
4.2 Limitations
The proposed API call graph malware hook detection approach assumes that the
model graph of malware samples might be from the same malware family if the calcu-
lated similarity value becomes high. However, more advanced kernel level malwares
use an effective obfuscation technique to evade detection which can affect the overall
effectiveness of the proposed approach. Next, polymorphic malware with advanced
packing poses a serious challenge when executing and extracting all its associated
parameters. Finally, few malwares can mimic the name of various operating system
resources; as a result, exploring the similarity value between two legitimate oper-
ating system resources is a challenging task. We point this issue as a possible research
direction.
5 Conclusions
Today, most malware authors have integrated API-based hooking method to avoid
detection from antivirus measures. This article presents a method that uses graph as
a tool to discover API hook-based attacks which are based on mistrustful system call
traces and its relationship among them. In turn, system calls are represented as a call
graph and comparing graph comparison is applied. Lastly, the system discovers the
similarity value to determine the presence of a malware. The experimental evaluation
results over the testing malware samples prove that our method incurs an average
12 P. Pandiaraja et al.
of 99% detection rate over the existing schemes. In addition, our method fabricates
better space complexity. In future, we plan to incorporate recent technology like
machine learning technique to automatically predict the occurrence of any attack
that targets exploiting system resources.
References
1. Current malware statistics. https://www.av-test.org/fileadmin/pdf/security_report/AV-TEST_S
ecurity_Report_2018-2019.pdf. Accessed on 2018/12/11
2. Bayer U, MilaniComparetti P, Hlauschek C, Kruegel C, Kirda E (2009) Scalable, behavior-
based malware clustering. In: Proceedings of the NDSS, pp 8–11
3. Willems C, Holz T, Freiling F (2007) Toward automated dynamic malware analysis using
CWSandbox. Secur Priv 2:32–39
4. Muthumanickam K, Ilavarasan E, Dwivedi SK (2013) A dynamic botnet detection model based
on behavior analysis. Int J Recent Trends Eng Technol 1:104–111
5. Muthumanickam K, Ilavarasan E (2014) Enhancing malware detection accuracy through graph
based model. Br J Math Comput Sci 4(15):2237–2250
6. Muthumanickam K, Ilavarasan E (2015) An effective method for protecting native API hook
attacks in user-mode. Res J Appl Sci Eng Technol 9(1):33–39
7. Muthumanickam K, Ilavarasan E (2015) COPDA: concealed process and service discovery
algorithm to reveal rootkit footprints. Malays J Comput Sci 28(1):1–15
8. Swiler LP, Phillips C, Ellis D, Chakerian S, Computer-attack graph generation tool. In:
Proceedings of the DARPA information survivability conference, pp 307–321
9. Guo H, Pang J, Zhang Y, Yue F, Zhao R (2001) HERO: a novel malware detection framework
based on binary translation. In: Proceedings of IEEE international conference ICIS, 2010, pp
411–415
10. Muthumanickam K, Ilavarasan E (2012) Automatic generation of P2P botnet network attack
graph. In: Das VV (ed) Proceedings of the third international conference on advances in
information on technology and engineering. Springer, New York, pp 288–293
11. Nath HV, Mehtre BM (2014) Static malware analysis using machine learning methods. In:
Martínez Pérez G, Thampi SM, Ko R, Shu L (eds) Proceedings of recent trends in computer
networks and distributed systems security: second international c onference, SNDS 2014,
Trivandrum, India. Springer, Berlin, 13–14 Mar 2014, pp 440–450
12. Oehmen CS, Peterson ES, Phillips AR, Curtis DS (2013) A biosequence-based approach to
software characterization. In: Proceedings of the IEEE international conference on intelligence
and security informatics, pp 330–332
13. Automated malware analysis—Cuckoo Sandbox. http://www.cuckoosandbox.org/. Accessed
2018/12/10
14. Pirscoveanu RS, Hansen SS, Larsen TMT, Stevanovic M, Pedersen JM, Czech A (2015)
Analysis of malware behavior: type classification using machine learning. In: Proceedings
of the international conference on cyber situational awareness, data analytics and assessment
(CyberSA), pp 1–7
15. Elhadi AAE, Maarof MA, Barry BIA, Hamza H (2014) Enhancing the detection of metamorphic
malware using call graphs. Comput Secur 46:62–78
16. Mehra V, Jain V, Uppal D (2015) DaCoMM: detection and classification of metamorphic
malware. In: Proceedings of the fifth international conference on communication systems and
network technologies, pp 668–673
17. Khodamoradi P, Fazlali M, Mardukhi F, Nosrati M (2016) Heuristic metamorphic malware
detection based on statistics of assembly instructions using classification algorithm. In:
Proceedings of the 18th CSI international symposium on computer architecture and digital
systems (CADS)
A Graph-Based Model for Discovering Host-Based Hook Attacks 13
18. Mosli R, Li, R, Yuan B, Pan Y (2016) Automated malware detection using artifacts in forensic
memory images. In: Proceedings of the IEEE symposium on technologies for homeland security
(HST), pp 1–6
19. Alabbas W, Al-Khateeb HM, Mansour A (2016) Arabic text classification methods: systematic
literature review of primary studies. In: Proceedings of the 4th IEEE international colloquium
on information science and technology (CiSt), Tangier, pp 361–367
20. O’kane P, Sezer S, McLaughlin K (2016) Detecting obfuscated malware using reduced Opcode
set and optimised runtime trace. Secur Inform 5(2):2–10
21. Salehi Z, Sami A, Ghiasi M (2017) MAAR: robust features to detect malicious activity based
on API calls, their arguments and return values. Int J Eng Appl Artif Intell 59(1):95–98
22. Bolton AD, Anderson-Cook CM (2017) APT malware static trace analysis through bigrams
and graph edit distance. Stat Anal Data Min 10(3):182–193
23. Muthumanickam K, Ilavarasan E (2015) Optimization of rootkit revealing system resources—a
game theoretic approach. J King Saud Univ Comput Inf Sci 27(4):386–392
24. Krishnan M, Egambaram L (2020) PAM: process authentication mechanism for protecting
system services against malicious code attacks. Sådhanå 45(141):1–12
E-Health Care Patient Information
Retrieval and Monitoring System Using
SVM
K. Sumathi and P. Pandiaraja
Abstract In healthcare, the modern technologies and the smart devices have brought
an excellent results. In Intensive Care Unit, these technologies brought a more facil-
ities to take care of patient health. The Internet of Things helps the gadgets to fit
with Internet. This provide a conjoin between the care taker and sick people which
leads to the duplex communication. The aim of the patient surveilling device is to
protect the patient in the intensive care unit. This system is used for analyzing the
patient essential movements and sends the report continuously to the doctor through
the help of the cloud. With the help of support vector machine algorithm, the data
get compared with available dataset. If the compared value gets reached above its
threshold value or below its threshold value the precaution message is send to the
server. The server sends the notification message to the care taker and provides guid-
ance for giving first aid. To collect these vital information, we need some sensors.
These sensors sense various body parameters such as the blood pressure, tempera-
ture (body heat), heart rate, and sugar level. In addition, our system also analyzes
the comma patient movement with the help of three-axis accelerometer sensor. The
heart rate is measured by using the pulse oximeter sensor; blood pressure is moni-
tored by blood pressure sensor. These sensors generate the report frequently. This
system mainly helps the patient by sending the message not only in emergency
case it also provides the precaution message if the value reached above or below its
threshold. The surveilling system helps the care taker and reduces the work pressure
and also this system overcomes the nursing staff’s shortage problem. The biomedical
data of the patient send through the server with the help of wireless communication
network and the data will be displayed on the mobile phone as well as laptop using
K. Sumathi
Department of Computer Science and Engineering, KSR Institute for Engineering and
Technology, Thiruchencode, TamilNadu, India
e-mail: thirusumathi83@gmail.com
P. Pandiaraja (B)
Department of Computer Science and Engineering, M.Kumarasamy College of Engineering,
Thalavapalayam, Karur 639113, TamilNadu, India
e-mail: sppandiaraja@gmail.com
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_2
15
16 K. Sumathi and P. Pandiaraja
web browser. This multitasking implementation in our system helps the health care
especially in Intensive Care Unit.
Keywords Microcontroller ·Internet of things ·Sensor network ·Three axis
accelerometer
1 Introduction
The patient surveilling device is used for observing the patient sensual signs. In
day-to-day life people are get suffered by many diseases. The hospitals are not even
having sufficient nursing staffs to monitor the patient especially in Intensive care
unit. The Internet of Things is the current growing technologies which helps the
people in medical field. It digitalizes the nursing care with the help of our device [1].
This device is connected with the sensor and if any symptoms are observed by the
sensor, it passes the message to the doctor with the help of server. This system is not
only specially made for any age group, it can be used by everyone who is admitted
in intensive care unit [2]. The wireless sensor networks are used for transferring data
from transceiver to receiver wirelessly. The main objective of the system is to act
as an intermediate in the situation, where the doctor is not available in the hospital
but even he or she can monitor their patient by getting details with the help of this
system [3, 4].
The Arduino microcontroller is used in which the sensors are getting connected.
The Arduino board continuously reads the input from the various sensors. It uses
the cloud database to store the data. The sensed information is transmitted to the
cloud with the help of Arduino [5, 6]. The GSM technology is used for location
tracking by which it can send the analyzed report of the patient to the doctor [7].
E-medicine plays a vital role in intensive care unit for the fast record of patient
details. The support vector machine algorithm is used for analyzing purpose where
the data sensed by the sensor is compared with the ideal data. When it identified
any problem, it automatically send message to the consultant person with the help
of the server, this saves the patient life [8]. The display devices are used in many
fields. But, particularly, it is very useful in medical fields. The display device must be
very accurate because of predicting disease [9]. Generally, LCD displays are used.
The resolution of image is very important. The high-resolution of ideal information
is efficient for the further treatment and diagnostics. The LCD display is composed
of constant number of pixels which helps to display the information on the screen
[10]. The LCD obtains the good quality image. Today’s scenario, most of the health
care units are built with audible alarm. The alarm is the spontaneous warning device
which helps in hospitals to alert and convey the message quickly and effectively.
Alarms in sensitive care unit that are enacted from many number of devices.
E-Health Care Patient Information Retrieval and Monitoring … 17
2 Related Works
The patient surveilling device is used to collect the data from the sick people in
intensive care unit with the help of Internet of Things. These information are sensed
through various sensors. Generally, the patient surveilling device helps to reduce the
works of care takers and it also saves the patient life [11]. This system works with the
help of IOT device like Arduino board or Raspberry Pi. The board consists of various
signals like analog or digital [12]. Each sensor connected with the Arduino board
to send the sensed information to the concerned person. The sensed information is
used to analyze for decision making and further to predict the diseases. The system is
classified into three stages. At first Stage the biosensor are used to predict the disease
[13].
The sensor like heart rate sensor, temperature sensor, and blood pressure sensor.
They are used to analyze patient daily status and send the sensed information to the
IOT device like Arduino or Raspberry Pi [14]. In this technical world, everything
was smart. In this smart world the smart devices like Arduino or Raspberry Pi are
interconnected with the objects around the environment. In medical field, Internet
of Things plays a major role, and it is more helpful to monitor and track. In second
stage, the sensed information is transmitted through server [15, 16]. To operate this
device, we need Wi-Fi connection. The Arduino board need to connect with the Wi-
Fi network using Wi-Fi module. Then only the information can be reached to the
doctor or the care taker [17]. Once the information is reached, it helps the care taker
or doctor for further treatment. At final stage, the system triggers alarm in case of
emergency situation. So that the care taker gets alert and pass the information to the
doctor. In this system, the main component is microcontroller board [18, 19].
The Arduino board consists of microchip, analog, and digital pins. It also consists
of USB port which makes us to connected with the laptop/pc’s. The board consists
of transceiver and receiver and also some light emitting diode. The Arduino uses
the USB port or external power supply to draw power automatically. In this system,
they applied support vector machine algorithm used for data classification [20]. The
patient-trained dataset is stored in the network, where the database gets updated.
The support vector machine algorithm checks the data’s threshold value; either it is
above or below; and during emergency case, it sends the message to care taker [21].
Then the messages pass through the mobile or display device like monitor/pc. The
display device mainly uses liquid crystal displays. Generally, the LCD is used for
high-resolution image processing [22]. If the resolution of the image is high, then
it is very helpful for the doctor to predict the disease very easily and the diagnostic
takes place in efficient way. The medical parameters value such as heart rate, blood
pressure and temperature are send to the care taker and doctor in digital form [23,
24]. So that they can analyze these values and monitor the patient in more precise
way in Fig. 1.
18 K. Sumathi and P. Pandiaraja
Fig. 1 Existing system for patient health monitoring
3 Proposed Model in IoT
3.1 Microcontroller
Esp8266 is called as a Wi-Fi module, but actually it is a microcontroller. Esp8266
is used to work with two ways. The first way is by using AT commands. Another
way is by using Arduino IDE. The AT commands is used to send the data from
Arduino to ESP. The Esp8266 module has eight pins which are used to perform
various functions. The maximum input voltage for Esp8266 is 3.3 V. If the input
voltage is greater than 3.3 V which causes damage to the module. Node MCU is an
open source LUA-based firmware. The development board of Node MCU V3 which
is used to run on Esp8266. The features of Node MCU are which has 4 MB flash
memory and 50 K usable RAM. The Node MCU consists of 30 pins. While 15 pins
are at the left and the other 15 pins are at the right. It has 16 pins for general purpose
input and output. Out of this, 16 pins for digital input and output 10 pins are used,
and 1 pin is used as the analog pin in Fig. 2.
Fig. 2 Node MCU with
ESP8266
E-Health Care Patient Information Retrieval and Monitoring … 19
3.2 Sensors
A sensor converts the impulses such as light, heat, sound, and motion into elec-
trical signals. These sensed information are gathered and sent to the interface which
converts them into a binary code then this binary information sent to the computer
for further process. There are two types of sensors; they are blood pressure and
temperature sensor.
3.3 Blood Pressure Sensor
Blood pressure is defined as pressure exerted by blood vessels while circulating the
blood. It is expressed in the ratio of systolic and diastolic pressure. Blood pressures are
measured by using sphygmomanometer but the blood pressure sensor itself measures
the artery without using mercury. In blood pressure sensors non-invasive method
is used to measure the blood pressure the normal blood pressure range is about
120/80 mmHg. When the range is above 180/120 mmHg means the person is in
serious condition.
3.4 Temperature Sensor
There are several types of temperature sensors they are thermocouples, resis-
tance temperature detectors, thermostats, infrared and semiconductor. To monitor
the human body temperature uses the thermocouples and resistance temperature
detectors. The normal body temperature for a person is about 37 °C.
3.5 Three Axis Accelerometer Sensors.
This three axis accelerometer is used to monitor the coma patients. This sensor
measures the acceleration of the body and compares the result with normal person.
3.6 Display Device
The display devices are used to collect the signals and display them on the monitor or
screen. In general the LCD displays are used in medical fields. Because the produce
good resolution of image and the doctor can easy to predict.
20 K. Sumathi and P. Pandiaraja
3.7 Alarm System
The alarm system is used to alert the people. Normally, the alarm system is used in
industries to alert the workers in case of emergency. This alarm now a day’s used
in many organizations and even in hospitals to protect the person’s life. In medical
field, the alarm is used to save the patient life by alerting the care taker. Normally,
in hospital, mild alert sound is used in order not to disturb the other person in the
hospitals.
3.8 SVM Algorithm
In this system, the support vector machine algorithm is used. The SVM algorithm
is a supervised learning algorithm. This algorithm is used to compare the data with
the help of hyper plane. The SVM works by mapping the data objects in the multi-
dimensional space. The SVM is classified into two types. The linear SVM is used to
draw a linear straight line, and it is used to find the difference between two classes.
In nonlinear, SVM we cannot use two dimensions to find the data’s. We need one
more dimension to identify the classes in Fig. 3.
Fig. 3 Support Vector Machine classifier algorithm
E-Health Care Patient Information Retrieval and Monitoring … 21
3.9 Cloud Server
Cloud server is considered to be a physical or virtual server the data’s get collected,
stored and hosted through the internet which runs on the cloud computing platform.
The cloud stores the data in the cloud storage in which the computer data’s and the
digital form of data are stored in the large logical pools. And, it is safe to store the
data’s in the cloud that can be easily retrieved or get accessed anywhere from anytime
through the internet server. It is impossible to delete all the data’s from the cloud. In
our surveilling device the sensed information get stored in the cloud. These collected
data are run through the cloud computing platform that send the notification or any
other messages to the particular doctor or the nursing staffs. With the help of web
portal address the sensed information get viewed.
4 Proposed Work
Our proposed system is used to protect the patient in intensive care unit. This system
helps the doctors and nursing staffs to reduce their stress level and also protect the
patient’s life. The surveilling device collects the patient blood pressure level, pulse
rate, and also body temperature with the help of sensors. The sensed information
from the sensor is connected to the node MCU Wi-Fi module. With the help of
the Wi-Fi module the collected data get compared with the dataset with the help of
support vector machine algorithm. The collected information get compared with the
available data and display the message through the help of cloud server. When the
compared data is above or below the threshold value, the notification is passed to
the doctors and nursing staffs. In normal system, it only pass the information during
emergency case. But in our system, we fix the threshold and pass the message during
emergency case and also pass precaution message. In addition, our system helps to
monitor the coma patient with the help of the sensor. The compared data pass on the
cloud which helps the doctor to access the report from any location and also it is very
helpful for the doctors to diagnosis the patient condition without reaching the health
care center in Fig. 4.
5 Results and Discussion
The modular system is used which helps to analyze the patient vital signs and also this
support vector machine algorithm works better to provide the results. To analyze the
body parameters by using various sensors which is connect to the cloud to diagnosing
the patients continuously in intensive care unit by monitoring through sensors and
also by using three-axis accelerometer sensors to monitor the coma patient move-
ment. The following figure shows the use case diagram of our proposed work and it is
22 K. Sumathi and P. Pandiaraja
Fig. 4 Proposed system for patient health monitoring
divided into three modules Analyzing body parameters through sensors, diagnosing
patient in intensive care unit and sending notification through cloud in Fig. 5.
The different wearable sensors are used to measure the coma patients body param-
eters like body temperature, muscle activity, pulse rate, and glucose level in the blood.
These tiny sensors are direct contact with skin of the patient and it can be used to
find the several diseases like fever, blood pressure, and sugar level. Then, numbers
of physiological parameters collected from these sensors are most preferred by the
doctors due to its accuracy.
A s mall hardware is used to preprocessing the acquired data and transmits desire
result to the other device through communication software. Normally, sensors are
small in size, light weight, and disconcerting mobility and movement of the patients.
The energy efficient components are used to operate the sensors and these compo-
nents may be working continuously without charging and replacement. The accurate
and secure recorded information of the coma patient in any location is reported to the
Fig. 5 Use case diagram for proposed model
E-Health Care Patient Information Retrieval and Monitoring … 23
Fig. 6 Analyzing body parameters through sensors
doctors using the data transmission system present in the communication software.
The result of this module is represented in the Fig. 6.
5.1 Diagnosing Patient in Intensive Care Unit
Remote monitoring of patients target several sub-groups of patients, such as patients
diagnosed with chronic illnesses, patients with mobility issues, or other disability,
post-surgery patients, neonates, and elderly patients. Automated health care services
are essential for our society and it reduces the burden of the nursing staff. The trans-
parency of this system increases the trust level of the patients. During the emergency
conditions, the buzzer and LED (Fig. 7b) present in alarm system alerts the doctors,
and she/he can act more quickly and handle the situation easily. The general steps in
diagnosing patient in ICU are represented in Fig. 7a.
5.2 Sending Notification Through Cloud
The real-world application challenges are solved by using the proxy-based approach
for end-to-end communication between the IoT-enabled living systems. It’s a chal-
lenge for large organizations to find cloud monitoring solutions [21–24] that provide
support in identifying emerging defects and troubleshooting them before they turn
into major issues. A sink node collects the signal from the sensor and forwards that
information to cloud via Wi-Fi or Bluetooth. The data stored in the cloud is further
processed whenever necessary. After processing the data and find out any emer-
gency then notification is send to the doctor using cloud enabled smart phone which
is depicted in Fig. 8.
24 K. Sumathi and P. Pandiaraja
Fig. 7 a Steps in diagnosing patient In ICU, b Diagnosing patient using LED display
5.3 Support Vector Machine
Support vector machine or SVM is one of the most popular supervised learning
algorithms, which is used for classification as well as regression problems. The goal
of the SVM algorithm is to create the best line or decision boundary that can segregate
n-dimensional space into classes so that we can easily put the new data point in the
correct category in the future.
This best decision boundary is called a hyper plane SVM chooses the extreme
points/vectors that help in creating the hyper plane. These extreme cases are called
E-Health Care Patient Information Retrieval and Monitoring … 25
HEALTH MONITOR
S.NO DETAILS DATE & TIME
1 T:23.93.75.07 HB:0 BP:94 15.02.2020 06:55:44 AM
2 T:25.39.75.70 HB:0 BP:94 15.02.2020 06:55:00 AM
3 T:24.41.75.95 HB:0 BP:94 15.02.2020 06:54:10 AM
4 T:33.69.92.64 HB:117 BP:171 15.02.2020 06:53:20 AM
5 T:23.93.75.07 HB:88 BP:171 15.02.2020 06:52:443AM
6 T:32.71.90.89 HB:120 BP:171 15.02.2020 06:52:07 AM
7 T:23.44.74.19 HB:120 BP:171 15.02.2020 06:51:11AM
8 T:26.86.80.34 HB:80 BP:123 15.02.2020 06:50:14 AM
9 T:23.93.75.07 HB:34 BP:92 15.02.2020 06:49:35 AM
10 T:30.76.87.37 HB:100 BP:157 15.02.2020 06:48:58 AM
11 T:25.39.77.70 HB:14 BP:94 15.02.2020 06:24:58 AM
12 T:23.44.74.19 HB:0 BP:94 T:23.44.74.19 HB:0 BP:94 15.02.2020 06:23:30 AM
13 T:25.39.77.70 HB:0 BP:94 T:25.39.77.70 HB:0 BP:94 15.02.2020 06:22:49 AM
14 T:24.90.76.82 HB:0 BP:94 T:24.90.76.82 HB:0 BP:94 15.02.2020 06:22:09 AM
15 T:24.41.75.95 HB:0 BP:94 T:24.41.75.95 HB:0 BP:94 15.02.2020 06:21:28 AM
16 T:24.90.76.82 HB:0 BP:94 T:24.90.76.82 HB:0 BP:94 15.02.2020 06:20:447AM
17 T:24.41.75.95 HB:19 BP:510 T:24.41.75.95 HB:19 BP:510 15.02.2020 06:20:06 AM
18 T:24.41 HB:0 BP:510 T:24.41 HB:0 BP:510 15.02.2020 06:02:01 AM
19 T:24.41 HB:0 BP:511 T:24.41 HB:0 BP:511 15.02.2020 06:01:05 AM
20 T:24.41 HB:0 BP:511 T:24.41 HB:0 BP:511 15.02.2020 06:00:25 AM
Fig. 8 Sending notification through cloud
as support vectors, and hence algorithm is termed as support vector machine. SVM
works by mapping data to a high-dimensional feature space so that data points can
be categorized, even when the data are not otherwise linearly separable. A separator
between the categories is found, and then the data are transformed in such a way that
the separator could be drawn as a hyper plane in Fig. 9.
The comparison of various approaches in naïve Bayes, decision tree, zero R, and
support vector machine for true and false classification approaches are mentioned in
Table 1 and its performance represents in Fig. 10.
Accuracy and precision of the different classification algorithms are calculated
by using the following formula
Accuracy =TP + TN
TP + TN + FP + FN (%)
Precision =TP
TP + FP (%)
The comparison of accuracy and precision of various approaches such as naïve
Bayes, decision tree, zero R, and support vector machine are mentioned in Table 2
and its results are represented in Fig. 11.
26 K. Sumathi and P. Pandiaraja
Fig. 9 Support Vector Machine decision boundary algorithm
Table 1 Comparison of true and false classification approaches
Approach Coma patient data (%)
True classification (%) False classification (%)
Naïve Bayes 84.13 15.87
Decision tree 67.89 32.11
Zero R 97.87 2.13
SVM 86.87 13.13
Fig. 10 Comparison of true
and false classification model
0.00%
20.00%
40.00%
60.00%
80.00%
100.00%
Naïve
Bayes
Decision
tree
Zero R SVM
Percentage of instances
Algorithms
True classificaon False classificaon
E-Health Care Patient Information Retrieval and Monitoring … 27
Table 2 Comparison of accuracy and precision of various approaches
Comparison of accuracy and precision
Approach TP FP Accuracy Precision
Naïve Bayes 0.833 0.339 92.411 71.075
Decision tree 0.962 0.019 98.137 98.063
Zero R 0.992 0.012 99.602 98.805
SVM 0.868 0.142 93.866 85.941
Fig. 11 Comparison of
accuracy and precision
0
20
40
60
80
100
120
Naïve
Bayes
Decision
tree
Zero R SVM
value
Algorithms
Accuracy
Precision
6 Conclusion
This system helps the ill people and also the doctors to detect the patient physiolog-
ical signs and provide the doctors the best report and reduces their work. Our system
overcomes the disadvantages of existing system. The support vector machine algo-
rithm is used which helps to analysis the patient vital signs and also this algorithm
works better to provide the results. The main idea of our system is we use the cloud
server to pass the data and also we use three axis accelerometer sensor to monitor the
coma patient movement. This system provides the efficient and good health services
to the patients. The feature of the system is to examine the patient from anywhere and
anytime. In our system we used future technologies and also we use various sensors
and it is easy to use.
References
1. Jaiswal S, Katake R, Kute B, Ranjane S, Mehetre PD (2016) Survey of health monitoring
management using internet of things (IOT). Int J Sci Res 15(11):243–2246
2. Sara GS, Sridharan D (2014) Routing in mobile wireless sensor network: a survey. Telecommun
Syst 57(1):51–79
3. Pustiek M, Beristain A, Kos A (2016) Challenges in wearable devices based pervasive wellbeing
monitoring. Int Conf Ident Inf Knowl Internet of Things, IIKI:236–243
28 K. Sumathi and P. Pandiaraja
4. Anusha N, Deepthi SM, Madhu M, Manish Adithya HM, Amutharaj J (2019) Smart healthcare
system assisted by IOT and emergency app. Int Res J Eng Technol l6(5):4551–4554
5. Kajaree D, Behera R (2017) A survey on healthcare monitoring system using body sensor
network. Int J Innov Res Comput Commun Eng l5(2):1302–1309
6. Meghana K, SudhirBabu A, KoteshwaraRao K, SreeLakshmi D (2019) Iot based patient health
monitoring system. J Inf Comput Sci 9(8):639–647
7. Pandiaraja P, Deepa N (2019) A novel data privacy-preserving protocol for multi-data users by
using genetic algorithm. J Soft Comput 23(18):8539–8553
8. Deepa N, Pandiaraja P (2019) Hybrid context aware recommendation system for E-health care
by merkle hash tree from cloud using evolutionary algorithm. J Soft Comput 24(10):7149–7161
9. Zhang Y, Liu H, Su X, Jiang P, Wei D (2015) Remote mobile health monitoring system based
on smart phone and browser/server structure. J Healthc Eng 6(4):717–738
10. Saranya M, Preethi R, Rupasri M, Veena S (2018) A survey on health monitoring system by
using IoT. IJRASET 6(3):778–782
11. Kishore KH, Nath KS, Krishna KV, Kumar DP, Manikanta V, Basha FN (2019) IOT based
smart health monitoring alert device. IJITEE l.8(6S):157–160
12. Vijayakumar P, Pandiaraja P, Balamurugan B, Karuppiah M (2019) A novel performance
enhancing task scheduling algorithm for cloud based e-health environment. Int J E-Health
Med Commun 10(2):102–117
13. Pandiaraja P, Deepa N (2020) E health care data privacy preserving efficient file retrieval from
the cloud service provider using attribute based file encryption. J Ambient Intell Humanized
Comput 12(5):4877–4887
14. Cirani S, Macro P (2015) Picone from wearable computing for the internet of things. IEEE
Comput Soc:35–41
15. Pandiaraja P, Vijayakumar P (2017) Efficient multi-keyword search over encrypted data in
untrusted cloud environment. In: Proceedings of the 2nd international conference on recent
trends and challenges in computational models (ICRTCCM ’17), pp 251–256
16. Mahalakshmi S, Latha R (2019) Artificial intelligence with the internet of things on healthcare
systems: A Survey IJATCSE 8(6):2847–2854
17. Shankar, A, Pandiaraja P, Sumathi K, Stephan T, Sharma P (2020) Privacy preserving E-voting
cloud system based on ID based encryption. J Peer-to-Peer Networking Appl. https://doi.org/
10.1007/s12083-020-00977-4
18. Sumathi K, Pandiaraja P (2020) Dynamic alternate buffer switching and congestion control in
wireless multimedia sensor networks. J Peer-to-Peer Networking Appl 13(6):2001–2010
19. Sravanan S, Abiramai T, Pandiaraja P (2018) Improve efficient keywords searching data
retrieval process in cloud server. In: International conference on intelligent computing and
communication for smart world (I2C2SW), pp 219–223
20. Rajesh Kanna P, Pandiaraja P (2019) An efficient sentiment analysis approach for product
review using turney algorithm. J Procedia Comput Sci 165:356–362
21. Hemasri M, Sumathi K (2019) SLA based combinatorial resource allocation model in cloud
computing. Int J Adv Sci Technol 29(7S):1151–1159
22. Sumathi K, JoseTriny K (2019) Secured data outsourcing in cloud with ECC encryption. Int J
Innov Technol Exploring Eng 8(8):1223–1227
23. Sumathi K, Naveena K, Prashanth P, Revanthkumar S, Srikeerthanaa AP (2019) E-Health based
patient surveilling device. Int J Emerg Trends Eng Res 8(3):792–796
24. Sumathi K, Adchaya P, Jayasri M, Nandhini B, PavithraJT (2019) Smart irrigation and agri-
culture monitoring system using cloud server based on IoT. Int J Adv Trends Comput Sci Eng
9(2):1082–1086
Number Plate Recognition Using Optical
Character Recognition (OCA)
and Connected Component Analysis
(CCA)
Puppala Ramya , Tummala Haswanth Chowdary, Pisupati Krishna Teja,
and Tadepally Hrushikesh
Abstract The number of automobiles has expanded dramatically during the last few
decades. As a result, tracking them became extremely difficult. In the event of a traffic
ticket or excessive speeding, identifying the automobile owner has become nearly
impossible. Image processing was used to identify car license plates to make this
practicable. That license plates will be retrieved from collected photographs utilizing
perception and computer vision algorithms, and then we will utilize OCR information
to recognize the license number. OCR stands for optical character recognition. We
employ cameras to record high-speed photos of number plates for image recognition,
and image processing algorithms to identify and validate the sequence of characters,
as well as to convert the number plate image to text. We now utilize number plate
recognition (NPR) to detect license plates. NPR is a computer vision technique that
enables equipment to scan license plates on automobiles swiftly and automatically
without the need for human intervention. Image processing methods include Hidden
Markov models, linear filtering, neural networks, and others. Our goal here is to
recognize license plates so that we can readily follow automobiles in the event of a
traffic penalty or excessive speeding.
Keywords Vehicle number plate ·Number plate recognition (NPR) ·Character
segmentation ·Recognized characters
1 Introduction
In recent years, number plate recognition or license plate recognition has proven to be
one of the most effective methods for vehicle surveillance. It can be used in a variety of
public locations for a variety of reasons, including traffic safety enforcement, car park
systems, and automatic vehicle parking systems. The four steps of an NPR algorithm
are as follows: (1) Image capture of a vehicle, (2) Identification of license plates, (3)
P. Ramya (B) · T. H . Cho wdar y · P. K . T e j a · T. Hrushikesh
Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation,
Guntur, India
e-mail: mothy274@kluniversity.in
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_3
29
30 P. Ra m y a e t a l .
Character segmentation, and (4) Recognition of characters. We will create software
for a real-time license plate recognition system in this project. Using OpenCV and
optical character recognition, this system recognizes and reads car license plates
automatically. It detects the license plate using OpenCV’s contour function. Lastly,
the license plate numbers are read using optical character recognition. Connected
component analysis was the method used to segment the images (CCA). Connected
regions denote that all of the pixels in the region are part of the same item. When
two pixels with the same value are adjacent to one other, they are said to be linked.
A license plate picture recognized in a car image is the first output obtained after
running the software. This is used as input for the next step, and CCA is used to bind
the characters in the plate using this image.
As seen in Fig. 1, the first task may appear simple, but it can be difficult to capture
a moving car in real time while making sure that all of its components, particularly
the license plate, are visible. Many algorithms today can recognize license plate
numbers in less than 50 ms. Identifying an NPR system’s effectiveness can be quite
important. Along with a visual and NPR quality assessment, a thorough study of
license plate identification is offered (LPR). The terms “number plate” and “license
plate” are used interchangeably in this literature. Each NPR is discussed in great
length in Sect. 2.
Fig. 1 Steps of number
plate recognition model
Number Plate Recognition Using Optical Character … 31
1.1 The Purpose of This Paper
Since it is impossible to differentiate between techniques, a variety of publications
built on the methods depicted in Fig. 1 are examined and categorized in accordance
with the methodology used in each approach. Since commercial products frequently
promise greater accuracy than is really achieved for promotional purposes, a survey
of them is not within the scope of this study. The following sections make up the
remaining text of this essay: A overview of several number plate detection tech-
niques is presented in Sect. 2. Section 3 discusses character segmentation t echniques,
whereas Sect. 4 discusses character recognition techniques (Figs. 2 and 3).
2 Detection of License Plates
Number plate recognition algorithms can be classified into more than one category
based on different methodologies. When identifying a vehicle number plate, the
following factors must be taken into consideration: (1). In a car image, the plate size
can change. (2). Anywhere in the car will have a licence plate. (3). Setting for a plate:
The colour of a licence plate’s backdrop may change based on the type of vehicle.
For instance, the background of a government vehicle number plate may differ from
that of other public cars. (4). A screw could be a character on a plate. The picture
segmentation approach can be used to extract a number plate. In diverse literature,
there are a variety of image segmentation methods. Color segmentation is used in
some plate segmentation methods. The following sections outline popular number
Fig. 2 Hardware setup for
NPR System
32 P. Ra m y a e t a l .
Fig. 3 Flowchart of the system
plate extraction methods, followed by a detailed examination of picture segmentation
techniques used in various NPR or LPR publications (Figs. 4 and 5).
2.1 Binarization of Images
The conversion of a picture to black and white is known as image binarization. This
approach uses a threshold to categorize pixels as black or white. The key issue,
however, is determining the appropriate threshold value for each image. Choosing
an optimal threshold value might be challenging, if not impossible, at times (Figs. 6,
7, 8, 9 and 10).
2.2 Detecting the Edges
For feature detection or extraction, edge detection could be a fundamental method.
In most cases, an object boundary with connected curves is the result of executing a
footing detection technique. Applying this method to complex photos becomes quite
difficult because it should lead to object boundaries with disconnected curves. Canny,
Canny-Deriche, Differential, Sobel, Prewitt, and Roberts crosses are a number of the
sting detection algorithms and operators that are employed.
Number Plate Recognition Using Optical Character … 33
Fig. 4 Images taken using a USB Camera
Fig. 5 Number plate extraction using smearing algorithm
2.3 Connected Component Analysis (CCA)
Blob extraction, also known as CCA, is a method for labeling subsets of related
components in a unique way using a heuristic. It s cans a binary image and identifies
pixels based on their connectivity conditions, such as the current pixel’s North-East,
North, Northwest, and West (8-connectivity). 4- Only the north and west neighbors
34 P. Ra m y a e t a l .
Fig. 6 Binary image
Fig. 7 Inverted binary image
Fig. 8 Line separation using row segmentation
of the current pixel are connected. The approach is more efficient and can be used
to perform automated picture analysis. This technique can be utilized for both plate
and character segmentation.
Number Plate Recognition Using Optical Character … 35
Fig. 9 Character separation using column segmentation
Fig. 10 Recognize character using OCR
2.4 Mathematical Morphology
Set theory, lattice theory, topology, and random functions are all used in mathematical
morphology. It is most typically applied to digital images, although it can also be
applied to other spatial structures. It was originally designed to process binary images,
but it was later expanded to handle grayscale functions and images. Erosion, dilation,
opening, and shutting are some of the basic operators (Table 1).
2.5 Related Work in the Number Plate Detection
Plate detection procedures such as those described in the preceding sections are preva-
lent. Aside from these methods, plate detecting methods have been discussed in the
literature. It is impossible to undertake a category-by-category analysis because most
of the strategies presented in this literature use multiple approaches. The sections
that follow discuss the various number plate segmentation algorithms. The sliding
concentric window (SCW) technique aims to identify the region of interest (ROI)
more quickly. From the upper left corner of the image, two concentric windows move
in two steps. To adapt camera distance and brightness under varied circumstances, the
36 P. Ra m y a e t a l .
Table 1 Some basic set operations
S. No Set operations
1Empty (Null) Set: ∅
2Subset of sets A and B: A ⊂ B
3Union of sets A and B: A ∪ B={xlx ∈ Aorx ∈ B}
4 Intersection of sets A and B: A ∩ B={xlx ∈ Aand x ∈ B}
5Disjoint/Mutually exclusive between sets A and B: A ∩ B =∅
6 Complement of a set A (with respect to a defined universe): AC = xlx /∈ A
7 Difference of sets A and B: A\B = A–B = xlx ∈ Aand x /∈ B
8Reflection (Transposition) of a set A: ˆ
A if A is symmetric
9 Translation of a set A by a vector z = (z1, z2): AZ = {xlx =a +z,a ∈ A}
Fig. 11 a A number plate with non-standard stylish font, b number plate with distorted angle, c
number plate with distorted angle
authors developed a revolutionary technique. Finding contours and related compo-
nents, selecting a rectangle region based on size and aspect ratio, initial learning
for adaptive camera distance/height, localization based on the histogram, gradient
processing, and closest mean classifier are some of the phases in the license plate
detection method. Once these steps are complete, the final detection result is sent for
tracking (Figs. 11, 12 and 13).
3 Character Segmentation
Characters are checked for the next step after locating the number plate. Character
segmentation can be done using a variety of approaches, just like plate segmentation.
It is impossible to discuss approaches by category because many fall into more
than one. This section discusses frequently related work in this field, followed by a
discussion. Some of the approaches outlined in Sect. 2, such as image binarization
and CCA, can also be used for character segmentation.
Number Plate Recognition Using Optical Character … 37
Fig. 12 Blurry number plate
Fig. 13 Number plates detected and recognized
For character segmentation, H. Erdinc Kocer used contrast extension, median
filtering, and blob coloring techniques. To make the image sharper, contrast exten-
sion is applied. Histogram equalization, according to H. Erdinc Kocer, is a popular
approach for improving the appearance of a low-contrast photograph. Unwanted
noisy regions are removed using median filtering. To detect closed and contact-less
zones, the blob coloring approach is applied to a binary image. By getting the connec-
tions into four directions from a zero-valued backdrop, this scanning procedure is
used to discover the independent areas (Fig. 14).
38 P. Ra m y a e t a l .
Fig. 14 Character segmentation
4 Character Recognition
The identification and creation of editable text from visual text is aided by char-
acter recognition, which is covered in more detail in Sect. 2. The bulk of number
plate identification algorithms employ a single character recognition approach. This
section goes into detail on each strategy. The optical character recognition (OCR)
tool is used by several algorithms to recognize characters. There is a wide range
of software available for OCR processing. Tesseract, a Google-maintained open-
source OCR tool with multilingual support, is one of the multilingual open-source
OCR tools. It is used to identify characters. The author tweaked it to get a character
recognition rate of 98.7%. In the Markov Random Fields (MRF) model of character
extraction, randomization is utilized to model the uncertainty in pixel assignment.
To maximize a posteriori probability, character extraction is done as an optimization
problem based on prior knowledge (Figs. 15 and 16)
Fig. 15 Recognition of characters
Number Plate Recognition Using Optical Character … 39
Fig. 16 Extraction of recognized characters from the image
5 Conclusion
As each step is dependent on the previous phase, it is obvious that NPR is a complex
method. It is currently impossible to achieve 100% overall accuracy because each step
is dependent on the previous phase. Different lighting conditions, car shadows, and
non-uniform license plate character sizes, as well as font and backdrop color, all have
an effect on NPR’s performance. Some systems are designed to perform just in these
limited circumstances, and they may not be accurate enough in other situations. Some
of the systems have been created and are being used in a given country. Just a small
number of NPR have been built for India. As a result, developing such a system for a
country like India has a lot of potentials. This paper presents a thorough examination
of recent advancements and potential trends in NPR, which will be useful to scholars
working on similar projects. In this project, we conclude that everyone must follow
the government’s rules, which require everyone to use a government of India number
plate. Nowadays, most people use the fancy number plates. This is a violation of the
government’s rules. So, our idea is that people with fancy license plates are scanned
by the camera and a challan is generated for the following person. Once this is put
into action, everyone will keep their government license plates and follow the rules
of the government.
Bibliography
1. Mitra D, Banerjee S (2016) Automatic number plate recognition system: a histogram-based
approach. IOSR J Electr Electron Eng 11:26–32
2. Chen W-K (2016) Linear networks and systems. Goyal BA, Bhatia R Various techniques for
number plate recognition-a review. Int J Comput Appl 143, Singh B, Kaur M, Singh D, Singh
G Automatic number plate recognition system by character position method. Int J Comput
Vision Rob 6(1–2):94–112
3. Qadri MT, Asif M (2009) Automatic number plate recognition system for vehicle identification
using optical character recognition. In: Proceedings of the 2009 international conference on
education technology and computer. Singapore, pp 335–338
40 P. Ra m y a e t a l .
4. Laroca R, Severo E, Zanlorenzi LA et al (2018) A robust real-time automatic license plate recog-
nition based on the YOLO detector. In; Proceedings of the 2018 international joint conference
on neural networks (IJCNN). Rio de Janeiro, Brazil, pp 1–10
5. Bhogale P, Save A, Jain V, Parekh S (2016) Vehicle license plate detection and recognition
system. Int J Comput Appl 137(9):31–34
6. Sharma G (2018) Performance analysis of vehicle number plate recognition system using
template matching techniques. J Inf Technol Softw Eng 8(2)
7. Council E (2010) Directive 2010/31/EU of the European Parliament and of the council of 19
May 2010 on the energy performance of buildings. Off J Eur Union 153:13–35
8. Sajjad K, Automatic License Plate Recognition Using Python and Opencv, Department of
Computer Science and Engineering, MES College of Engineering, Kuttippuram, Kerala, 2010.
Muñoz J, Craft M, Ahmadian M, Wrobel S (2013) Multifunction LIDAR sensors for non-
contact speed measurement in rail vehicles: Part I—system installation and implementation.
In: Proceedings of the 2013 joint rail conference. Philadelphia, PA, USA
9. Chang S-L, Chen L-S, Chung Y-C, Chen S-W (2004) Automatic license plate recognition.
IEEE Trans Intell Transp Syst 5(1):42–53
10. Agarwal P, Chopra K, Kashif M, Kumari V (2018) Implementing ALPR for detection of traffic
violations: a step towards sustainability. Procedia Comput Sci 132:738–743, 2018. Vehicle
number plate detection. In Soft computing: theories and applications, pp 453–461, Springer,
Berlin, Germany, 2019.
Cartoonify an Image with OpenCV Using
Python
Puppala Ramya , Penki Ganesh, Kopanathi Mouli,
and Vutla Naga Sai Akhil
Abstract This article describes a technique for generating cartoon-like images from
digital pictures. The method used today is different from how things were done in
the past. This study focuses on the various tactics used during the process that, when
used layer by layer, provide a product that is well balanced. We usually research how
to combine several functions in a specific way to provide a filtered and composite
outcome. Various functions’ mathematical foundations and mechanisms have also
been discussed. This article provides examples of a variety of cartooning techniques.
Any of the methods given here can be used to turn any type of obtained photograph
into a cartoon, including pictures of people, mountains, trees, flora and fauna, etc.
Keywords Use filters to cartoonize images in Python ·Including the bilateral ·
Gaussian ·Pencil edge ·Pencil sketch ·Laplacian ·Median filters ·Computer
vision
1 Introduction
Cartoons are pictures of fictional or based on real-people figures. These days, semi-
realistic or non-realistic paintings that satirically or humorously reflect a situation or
an event are popular. One of the earliest instances of traditionally animated movies is
Fantasmago-rie (1908), in which every frame was drawn by hand. This tradition of
hand-drawn animation frames is still prevalent today. Walt Disney caught up to the
competition with their excellent animated series and improved animation cartoons
to a new level. Cartoonists used to hand-draw these cartoons in the past, but as
“Anime” gained popularity, it became more challenging for them to do so because
it took a lot of time and could not be undone if a mistake was made. With the
development of technology, a wide range of software for digitally designing pictures
was created, reducing the need for human labor and speeding up the process artists
P. Ramya (B) · P. Gan e s h · K. Mouli · V. N. S. Akhil
Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation,
Guntur, India
e-mail: mothy274@kluniversity.in
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_4
41
42 P. Ra m y a e t a l .
images compared with, it is more efficient. Which got better over time by getting
more features added. Toy Story, the first fully computer-animated feature film, was a
huge success in 1995. It featured interactively stunning characters, and the incredible
animation brought them to life.
2 Objectives
2.1 Filter by Median
It is a nonlinear filtering method for reducing noise and increasing edge identification
accuracy in a picture. Additionally, consider the image’s edges when removing noise
from it. Exceptional outliers that would distort the average.
2.2 Filter Laplacian
A Laplacian filter is an edge detector that determines an image’s second derivatives by
detecting the rate at which the first derivative is changing. This establishes whether
a change in the values of nearby pixels is the result of an edge or is a result of a
continuous progression. A linear differential problem is treated using the Laplacian
approach using a second-order derivative.
2.3 Filter by Median
It is a non-linear filtering method for reducing noise and increasing edge identification
accuracy in a picture. Additionally, consider the image’s edges when removing noise
from it. The best candidates for median filters are extreme outliers that might distort
the average. The median filter is frequently used to eliminate noise, salt, and pepper
(Fig. 1).
3 Implementation
3.1 Step by Step Implementation
Main system Purpose is to turn on. In this function, a sidebar is present. You may
simply build one using the sidebar choose box function in Streamlit. (1) The sidebar
has some values, each of which is associated with a function. (2) When a user clicks
Cartoonify an Image with OpenCV Using Python 43
Fig. 1 Original, pencil sketch, pencil edge images
on one of them, the associated procedure is launched. By default, the Pencil Sketch
string is chosen, and the “if” condition calls the PencilSketch() function with st.title
to create a bold title. Additionally, we can utilise St.Image to display any image on
our Streamlit app.
After that, the section with the image-editing tools appears. These include the bilat-
eral filter, detail enhancement, and pencil sketch. The st.slider method in Streamlit is
used to generate an interactive slider. Here, a widget that lets users choose their own
photos from their local system is added using the streamlit.file uploader() function. By
browsing or dragging and dropping the image into the box-covered area surrounding
the button, users of this widget can choose their image. Text can be added to the app,
such as messages or any important information, using the streamlit.write() func-
tion. The image is displayed after the user has completed selecting it by calling the
streamlit.image() function with the necessary inputs (Figs. 2, 3 and 4).
Fig. 2 Bilateral
44 P. Ra m y a e t a l .
Fig. 3 Detail enhancement
Fig. 4 Original image
3.2 Statement of the Problem
How a photograph is cartoonized depends on the strategy used by the algorithm. There
are numerous ways to carry out the same task. The most widely utilized techniques
are generative Adversarial networks (GAN), a machine learning framework that
creates new data based on training data, and the OpenCV library, which we use on
our system. (1) The existing system is based on the filters that were applied to the
input image and the OpenCV library. To construct a custom filter, a variety of filters
can be combined or used separately. Among the most famous and well-liked filters
are Medi-anBlur(), GaussianBlur(), Laplacian(), BilateralFilter(), and many others.
The best results are obtained when these filters are combined. utilizing a single,
minimally capable filter.
Cartoonify an Image with OpenCV Using Python 45
3.3 Advanced Technical Approach
Image processing includes the critical step of filtering an image. Among other things,
it can be used to eliminate blur, noise, and detect edges. Algorithms both linear and
non-linear are used for filtering. The appropriate filter should be used for every
particular objective. If the input image has a large magnitude and there is little to
no noise, the filter is non-linear. If the input image is low magnitude and noisy, it is
sometimes referred to as a linear filter. Due of their simplicity and speed, linear filters
are the most popular filters. The Gaussian and Laplacian algorithms are used by linear
filters, and the median and bilateral methods are used by nonlinear filters. Algorithms
used include pencil sketch, detail enhancement, pencil edge, and bilateral.
After that, the section with the image-editing tools appears. These include the bilat-
eral filter, detail enhancement, and pencil sketch. The st.slider method in Streamlit is
used to generate an interactive slider. Here, a widget that lets users choose their own
photos from their local system is added using the streamlit.file uploader() function. By
browsing or dragging and dropping the image into the box-covered area surrounding
the button, users of this widget can choose their image. Text can be added to the app,
such as messages or any important information, using the streamlit.write() func-
tion. The image is displayed after the user has completed selecting it by calling the
streamlit.image() function with the necessary inputs (Fig. 5).
In this work, we suggested an image cartoonization online software that transforms
real-world photographs into exquisite cartoon-style images. Because the animation
industry isn’t going anywhere and the requirements are getting higher by the day, this
technique enables for the addition of features and is easily convertible to any other
source code needed for larger modules. The image will be less pixelated as a result of
Fig. 5 Pencil sketch image
46 P. Ra m y a e t a l .
Fig. 6 Process diagram
the automation system patching up our technology. (1) There have been four filters
applied: pencil Sketch—Creates a pencil sketch from the contents of an image. (2)
Smooth the image by reducing noise while preserving the edges with a bilateral filter.
(3) Detailed enhancement improves the details by sharpening the image and adding
detailed noise. (4) Overall, the system is successful because it produces satisfactory
results with a large number of photographs, and support will increase as we continue
to improve the system to produce the best results. (5) Pencil edge—Converts the
image into one with only the most important edges and white fills the insides.
There are many techniques that have been developed to create flat-shaded images
that look cartoonish. In pencil sketch, we applied the default sigma values filter
and the default Gaussian blur technique with 25 × 25 pixels to blur our image. By
increasing the filter size, which is also used to lessen image noise, we may create
fine lines for our sketch. A component of the Xpdf software suite is pdftotext. The
Xpdf-based Poppler program also offers a pdftotext implementation. The majority of
Linux systems, poppler-utils includes pdftotext. La-placian filter kernels frequently
have negative values cantered inside the array in a cross pattern. Either a 0 or a 1
number applies to the corners. The centre value may be either positive or negative.
In the following array, a 3 × 3 kernel for a Laplacian filter is displayed. The first
test makes use of a landscape photograph. Although there are numerous objects in
the picture, the way the buildings are arranged and how the horizon is formed, as
well as the leading lines, give the picture a very appealing and lively appearance.
Since the filters work well with concrete items, the technique produces results that
are considerably cleaner and more “cartoonified” when used with cityscape images
(Fig. 6).
4 Proposed System
The goal of this project is to create an intuitive application that enables users to apply
cartoon filters to whatever photos they want. On a wide range of photographs, the
Cartoonify an Image with OpenCV Using Python 47
filters are designed to produce beautiful and amusing outcomes. Python was used to
develop the code, therefore, the system emphasizes the program’s simplicity. Python
is regarded as the language that is the most “fun” and simple to learn, with a wide range
of applications in every field. Anyone of any age may utilize the interface, regardless
of system or service provider. It can be accessible from any device with a browser
and an internet connection when it is hosted online. There are some considerations
when using the detail enhancement filter. The subject must be distinguished from
the background in a shot to be considered ideal, and the lighting and metadata must
all be well distributed and not overdone. Use the slider carefully because even the
smallest adjustments can have a significant impact on the result. The program that
accurately records even the smallest information on the edge. The filter allows for
adjustment of the edge detecting power. However, bear in mind that the higher the
power, the more unnecessary edges will show while adjusting the filter.
4.1 Challenges and Problem
Network training for different image types is time-consuming and computationally
intensive (GPUs). Style variations may occur depending on the image’s content. The
type of given content image completely determines how accurate the cartoon-like
appearance will be (Figs. 7, 8 and 9).
5 Conclusion
As a consequence, we were able to show how a cartoon may be created from an
image. Examples of how an image is turned into a cartoon are offered. The hard-
ware and software requirements for converting images to cartoons are also shown in
this document. In this study, a stunning diagram illustrates the methodical process
of converting images to cartoons, along with the related equations and algorithm.
Additionally, we’ve listed a few of the difficulties and problems that might come up
when cartoonizing a captured image. In this study, we also looked at the significance
and degree of cartoonizing the content image (Fig. 10).
48 P. Ra m y a e t a l .
Fig. 7 Pencil edge, detail enhancement, bilateral images
Fig. 8 Original image
Cartoonify an Image with OpenCV Using Python 49
Fig. 9 Bilateral image
Fig. 10 Image deenhancement
References
1. Gatys LA, Ecker AS, Bethge M (2016) A neural algorithm of artistic style
2. Gatys LA, Ecker AS, Bethge M (2016) Image style transfer using convolutional neural networks
3. Johnson J, Alahi A, Fei-Fei L (2016) Perceptual losses for real-time style transfer and super-
resolution
50 P. Ra m y a e t a l .
4. Li C, Wand M (2016) Precomputed real-time texture synthesis with markovian generative
adversarial networks
5. Ulyanov D, Lebedev V, Vedaldi A, Lempitsky V (2016) Texture networks: feed-forward
synthesis of textures and stylized images
6. Li Y, Wang N, Liu J, Hou X (2017) Demystifying neural style transfer
7. Dumoulin V, Shlens J, Kudlur M A learned representation for artistic style, 2017 Summary
We used computer vision algorithms to turn a typical image into a comic in the preceding
demonstration. We’re going to have a lot of fun with computer vision techniques. The cartoonie
image function is then called after we check what we pressed on the keyboard. The sketch mode
attribute has distinct values in the two calls, resulting in two different outputs (we mentioned
what the output will look like earlier in this post)
Web Design as an Important Factor
in the Success of a Website
Puppala Ramya , K. Jai Sai Chaitanya, S. K. Fardeen, and G. Prabhakar
Abstract The internet has grown as a new business medium with e-commerce in
recent years, where in a good web design plays an important role. Hence, a detailed
study has been taken up to understand the characteristic features of a good web
design for the successful e-commerce websites (Journal of Systems and Information
Technology, Volume 11, Issue 2 (2009–05-03)). It has been identified that along with
the outer appearance of website, focus should be upon the usability of a website by
all kinds of users including the visually challenged. This article provides a few tips to
make it convenient for the low sight or no sight people using tools like screen readers
and voice synthesizers based on the case studies conducted. However, there is no
optimum design specified, as the goods sold are different along with the consumers
of different geographic locations. Applying accessibility and usability standards from
the beginning of the design phase is far less expensive than integrating them later
(Maria Claudia Buzzi, Marina Buzzi, Barbara Leporini. Chap. 4 Accessibility and
usability of web content and applications, IGI Global, 2010). In the long term, making
things more accessible and inclusive is a win–win situation and will increase any
individual’s and organization’s overall efficiency and effectiveness of engagement.
Keywords Web design ·e-commerce ·Visually challenged ·Consumer happiness
1 Introduction
The Internet has grown rapidly in popularity as a new business medium in recent
years. Throughout the world, there are more than 60–70 million websites and nearly
1400 million target audience providing practically limitless opportunities in the
market. As a consequence, there has been a major increase in competency and organi-
zations are questioning how to get the best results. Understanding what people want
might be the first step towards finding a solution. As a result, a considerable body
P. Ramya (B) · K. Jai Sai Chaitanya · S. K. Fardeen · G. Prabhakar
Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation,
Guntur, India
e-mail: mothy274@kluniversity.in
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_5
51
52 P. Ram y a e t al .
of research has evolved concentrating on the factors that impact an e-commerce
website’s performance from the perspective of consumers. A good web design is
an important factor in the successful organizations of e-commerce in the present
competitive world (Figs. 1 and 2).
This study focuses on a broad examination of views within the marketing profes-
sion. The implementation of successful interfaces that produced favorable responses
from users has sparked these study lines. In order to obtain good website satisfac-
tion or to increase the consumer’s online purchase intentions, a good web design in
essential. A successful web design includes following contents:
Fig. 1 Website contents
Fig. 2 Website offers
Web Design as an Important Factor in the Success of a Website 53
Despite the identification of the importance of web design in the creation of
successful virtual stores, there appears to be a surprising lack of content in the
literature on how to handle the many parts of website design. Hence, in this paper,
we would like to answer the following research questions:
RQ1: From a consumer’s perspective, what are the most important aspects that
influence the success of e-commerce websites? What function does site design play
in this? RQ2: What are the primary characteristics of effective virtual shop web
design practices?
2 Affective Factors for a Successful Website of e-commerce
In recent times, many studies have attempted to determine which aspects contribute
to a website’s success. In this regard, marketing literature has emphasized the
consumers’ perspective in defining how a successful e-commerce website should
be. As Carlos Flavian et al. [1] quoted in their article “A successful website attracts
customers, makes them believe the site is trustworthy, dependable, and reliable, and
creates customer pleasure”.
Following this approach, some authors have emphasized customers’ thoughts and
perceptions of the value provided by website qualities, while others have highlighted
the primary elements of website quality from a consumer’s perspective. Furthermore,
various study lines have been produced with the goal of highlighting those decisive
aspects and underlining the significance of obtaining online consumer happiness and
its influence on his purchase intention. In this regard, it is worthful to mention what
Factors might impact the consumer’s opinion against buying on the internet, and
their intentions to buy on the internet. Consumer’s attitudes and purchase intentions
are influenced by elements such as—product perceptions, shopping convenience,
appropriate information and appealing look, price details, etc.,
To analyze the performance of websites, we concentrated on content and design
to satisfy consumer perceptions of the structure, image and style of websites, along
with transaction information and navigation. The findings revealed how important it
is to include user perceptions when studying websites. However, the external factors
such as the source, industry, and the scale of the website have also a key role in
success.
According to the findings the most important characteristics of e-commerce
websites in order to enhance online shoppers’ purchasing intentions are—secu-
rity and the protection of personal data, the quality of the visual design, and the
requirement of providing relevant navigation to meet their needs (Fig. 3).
As a result, the significance of showing high-quality information, strong content
and an effective and appealing navigation system becomes clear. These are the most
important advantages of using an e-commerce website.
54 P. Ram y a e t al .
Fig. 3 Chart of e-commerce website
3 Importance of Webdesign in e-commerce
A successful web design includes a coherent and useful form structure which is
planned to meet the purpose of the consumer’s search. It should be designed in
an artistic way to attract the various perspectives of the consumers, to excite their
emotions and to increase their on line purchasing intentions.
While designing a website, we should keep in mind the various aspects of its
usability such as easy access in navigation, easy to memorize the main operations,
simple to look at, avoiding mistakes in operation and enhancing pleasure, and finally
the consumer satisfaction.
Usability may also be thought of as a technique for assessing the quality of a
website in this vein. As a result, a system’s simplicity of use might lead to more
concept learning and a better capacity to predict how that system performs. Usability,
in particular, enhances the consumer’s grasp of the content and activities that must be
understood in order to attain a goal (e.g., to place an order). This lowers the chances
of making a mistake and raises trust levels (Figs. 4 and 5).
In terms of website design elements, a good web design must give not only
beauty and attractiveness, but also high levels of usefulness, as it impacts the user’s
emotional states. As a result, a well-designed website should have excellent usability.
The usage of a website can be made more pleasurable by a pleasing design. In reality,
Web Design as an Important Factor in the Success of a Website 55
Fig. 4 Examples of successful website
Fig. 5 Visual of website
a high level of perceived usability may lead better satisfaction, trust, and loyalty to
a website. The verified measurements of a website’s usability and design are: iden-
tifying characteristics such as response time (download delay), content arrangement
(navigation), and the website’s information and content. Media richness elements,
such as a website’s ability to change its appearance and contents (interactive) and
the availability of feedback between the seller and the user, were also identified to
be drivers of a website’s success in the study.
56 P. Ram y a e t al .
The study also found that the aesthetics of a website’s design had a significant
impact on its success. The impact of the web environment on consumers was studied,
and it was revealed that the environment’s insights influence the consumer’s cognitive
and affective states, as well as their purchasing behavior toward the product. The
look of the website is an important aspect in improving information perception,
which allows individuals to do better cognitive mapping and evaluations of decision-
making. It can be explicitly claimed that graphical representations such as symbols,
colors, photos, and animations offer websites more vibrancy. This information may
increase people’s pleasure with the website and their navigational experiences.
As a result, a large portion of the literature emphasizes the importance of factors
such as acceptable appearance, simplicity of navigation, convenience of use, security
and privacy, and content information. These variables influence customer behavior.
In an e-commerce context, website success is critical. The lack of consensus in the
studies on how those factors should be addressed appears to be unusual. Therefore,
it appears that a set of principles should be defined in order to construct interface
that satisfies both users as well as business requirements.
4 Usability
Usability is a quantitative, or measurable, statement of how easy it is for users to do
activities for which a webdesign is created. As presented in the article by Lisa (et al.
2014) Usability is defined as, “the efficacy, efficiency, and happiness with which
specific users achieve specific goals in specific situations.” [2] Correspondingly, the
factors that determine a site’s usability must include error-free location of informa-
tion, e-commerce transactions with ease, user/customer satisfaction, remembering
the organization of site, and other functionalities.
The effectiveness and efficiency as defined by D. Reddig (et al.) at 2008 conference
is as follows:
• “The precision and completeness with which certain users may attain specified
goals in specific circumstances” is characterized as “effectiveness”.
• “The resources invested in proportion to the precision and completeness of goals
attained” is how efficiency is defined.
• Navigability and interactivity are also the important aspects for website usability.
• The other important factors that impact the novice users are efficiency, errors,
satisfaction, easily memorable activities, and learnability.
• Natural content grouping, presentation, and control uniformity throughout the site
with clear and meaningful labels; contextual navigation like where and how can
he get the product desired, are also the other factors the impact the users.
• Care should also be taken in terms of page arrangement. Instead of loading too
many items on a page, related items can be grouped to attract user interest.
Web Design as an Important Factor in the Success of a Website 57
5 A Case Study of Totally Blind Persons Interacting
on the Internet
Persons who are fully blind have more difficulty undertaking specific tasks than
people who have other sensory abnormalities such as poor vision, motor, or hearing
impairments [3]. Petrie (et al.), published the results of accessibility testing of 100
websites with users with visual, motor, and perceptual problems, indicating that
websites that are accessible for persons with varied abilities may still be aestheti-
cally appealing. In total, 100 websites from five different businesses were evaluated
using automated verification and user testing, with 51 persons of various abilities,
including 10 fully blind people. The average task success rate was 76%, but when
only the completely blind were included, it dropped to 53 percent (the lowest score of
all the user categories) [4]. Similarly, the authors discovered that the blind had more
difficulty with user satisfaction than other differently-abled users [5]. (4.2on a 1.0.7
Likertscale, the lowest score of all the user groups). Researchers from Manchester
Metropolitan University evaluated a group of blind and visually impaired users who
undertook four information-seeking tasks, including the use of search engines, in
order to highlight non-visual access issues. Visually challenged persons spend 2.5
times longer than sighted users to search the Internet for a specific piece of informa-
tion. When given a set of assignments, blind respondents took twice as long as sighted
users to analyze search results and three times as long to browse the linked internet
sites [6]. The three non-technical requirements in WCAG 1.0 have been developed
into nine particular concepts for designing better text user interfaces (ETI). The
authors Maria Claudia Buzzi et al. [4] demonstrated that the ETI guidelines improve
usability by evaluating the efficiency, errors, and user satisfaction of a web user inter-
face developed according to these specific guidelines regarding GUI conformance
to standard; the study involved 39 blind users who were asked to complete two tasks
[4]. We shall concentrate our discussion in the following section of this chapter on
visually impaired & especially on the needs of the completely blind, because the
authors are specialists at these.
A screen reader and a voice synthesizer are the required software for blind people
who use computers which help them to scan the information on the given website and
vocalizes the particular information as required. Visually challenged find it easier
to navigate using the arrow keys, tab keys, and other access keys using specially
designed keyboard than the mouse pointers for scrolling and pointing, selecting,
etc. They found it more advantageous using voice synthesizer to give and receive
instructions and to browse the content on a web site. So utmost care should be taken
while designing a web site to make it more user friendly using the keyboard access.
Users may have navigational challenges even when websites fulfill accessibility
rules. In general, a web page may include a variety of access options such as tables,
drop down menus for extra information or multiple options, and so on. This may
create a problem for the challenged persons to access the website where in they
make use of special keyboards, screen readers, voice synthesizers, etc. [2] (Fig. 6).
58 P. Ram y a e t al .
Fig. 6 Pie chart technologies preferred by blind users
• Overloaded information may prove hazardous to the challenged persons as they
have to stop and start the screen readers frequently.
• While browsing via screen reader, the user may miss the general context as the
screen reader can read only bits of information.
• The text connected with a link will appear on the Braille display or the synthesizer
will play it (e.g., “.PDF,” “additional information,” and so on.) However, she or
he is unaware of what is written before and after [4]. As a result, it’s possible that
the reading procedure may need to be repeated.
• Mixing up the content and structure also proves to be an issue for the blind
reader. As they arrive in the code, the screen reader announces the most significant
interface components such as links, graphics, and window objects. These elements
are critical to understand the page structure. However, the actual reading process
might be taxing for the user, necessitating a lot of mental work [7].
• As UI aspects are difficult to comprehend, the links, content, and button labels
should be self-explanatory and context-independent.
• Finally, visual content cannot be accessed by a blind person in general. As a result
the various elements of visual content like captioning, and video conferencing may
be provided with extra tools like audio links, etc., for the effective communication
of information.
• User interface architecture and organization are critical for users working with
assistive technology. Because websites (or software windows) are built for visual
engagement, it is challenging for visually challenged individuals to navigate the
Web.
Developers should be mindful of how material is seen when dealing with a screen
reader in order to build user interfaces effectively to address or lessen the problems
by focusing on the page layout. Care should be taken to see that the information
may be easily accessed with screen reader. Additional information may be given
where in photos are displayed for the general users to suit the needs of the visually
Web Design as an Important Factor in the Success of a Website 59
challenged. Every item should be properly labeled for easy navigation. The ability
of a blind user to navigate a website by breaking it down into logical pieces may
improve their experience in two ways: it gives a page overview and allows them to
go from section to section. Using proper heading levels helps the visually challenged
with the navigation as the screen readers have particular instructions while changing
from one heading to the other.
6 Conclusion
The rapid growth of the Internet in recent years has been accompanied by an intensely
competitive environment. This study focuses on specialist literature and empirical
data on the primary characteristics that influence a company’s level of success in
internet commerce. In particular it is possible to emphasize factors connected to
website design.
The First research question from this study was to identify some primary charac-
teristics that influence the success of e-commerce websites from the perspective of
consumers. The literature studied gives us a conclusion that web design is an impor-
tant aspect in achieving favorable results, since it has an impact on users, online
customer’s perceptions and behavior’s. As a result, website design is an excellent
foundation for online businesses to build customer happiness, confidence, and posi-
tive intentions toward the website. We’ve focused on the interaction between usability
and web design in particular to help all kinds of users to navigate the website, giving
them control over their own tasks and a sense of freedom. As a result, web design
plays a crucial role in the success of a website. In addition, we have also high-
lighted a number of examples of good design strategies in order to identify the key
characteristics of a successful web design in the digital store.
Nonetheless, it is reasonable to assert that there is no one optimum design because
it differs based on the type of- goods being sold, person in front of the computer,
and the geographic location from which the website is accessible. Design investment
is necessary to instill confidence in consumers’ thoughts, resulting in greater online
purchase intentions. Furthermore, privacy and security must be taken into account at
all times and in all areas of the website. In order to establish the crucial components
for reaching high levels of online firm success, this study proposes a Decalogue in
this area. As a consequence, we’ve created these set of principles that can assist
e-commerce websites seem better. This Decalogue might be quite useful for web
designers in establishing the most important factors to consider while developing
a website. We’ve discovered that online users’ viewpoints must be emphasized in
every part of the website design; as a consequence, the dimensions outlined in this
set of suggestions serve as the foundation for influencing online users’ perceptions
and behavior’s, and defining the success of a website.
To begin, it seems reasonable to evaluate the visual attractiveness of the website
as it will influence the consumer’s relationship with the firm. Secondly, the user’s
60 P. Ram y a e t al .
ability to navigate through the website is focused upon which increases the site’s
ease of use and emotions of freedom while navigating.
Finally, e-business must carefully manage the information and contents of their
websites, providing high-quality data in appropriate formats. Designers should
attempt to homogenize the steps of the buying process so that people can better
understand and comprehend the commercial process on the Internet and feel more
secure about purchasing a product. Many web designers have been led astray by tech-
nical issues than by user needs, resulting in the creation of more complex websites.
Despite the fact that these technologies have an influence, it will always be neces-
sary to follow basic design principles, such as creating a beautiful atmosphere for
the clients to give the website a trustworthy design and experience to make it easy
to use and explore.
To conclude, this chapter examined a number of important issues for ensuring that
programs, such as everyone has access to the Internet, work properly. The concepts of
assistive technology, accessibility, and usability were studied in connection to each
other after a review of the many types of disabilities and their impacts and can be
used by a diverse group of individuals, regardless of their age or skill.
References
1. Flavián C, Gurrea R, Orús C (2009) The effect of product presentation mode on the perceived
content and continent quality of web sites. Online Inf Rev
2. Leporini B (2008) Evaluating a modified Google user interface via screen reader Univ Access
Inf Soc 09/2008
3. Akhter F, Buzzi MC, Buzzi M, Leporini B (2009) Conceptual framework: how to engineer
online trust for disabled users. In: 2009 IEEE/WIC/ACM international joint conference on web
intelligence and intelligent agent technology
4. Buzzi MC, Buzzi M, Leporini B (2010) Accessibility and usability of web content and
applications, Chap. 4, IGI Global
5. Buzzi MC, Buzzi M, Leporini B, Akhter F(2010) Is Facebook really "open"
to all?”. In: 2010 IEEE international symposium on technology and society
6. Andronico P, Buzzi M, Leporini B, Castillo C (2006) Testing google interfaces modified for the
blind. In: Proceedings of the 15th international conference on World Wide Web—WWW 06,
2006
7. Leporini B (2007) Learning by e-Learning: breaking down barriers and creating opportunities
for the visually-impaired. In: Lecture notes in computer science
8. Witt H (2008) Human-computer interaction, Chap. 4. Springer Science and Business Media
LLC
Earlier Selection of Routes for Data
Transfer In Both Wired and Wireless
Networks
S. NagaMallik Raj , S. Neeraja , N. Thirupathi Rao ,
and Debnath Bhattacharyya
Abstract In both wired and wireless communication networks, the transfer of huge
data was the major issue. To solve this issue, we are in the process of working with
both admission control and control transfer mechanisms such that to transfer huge
amount of data in the current networks without disturbing the regular flow of data
in the networks. A scheduling algorithm was also in the process to achieve this goal
of transfer of substance amount of data from source to desti nation securely. In the
current system, for every transfer of bulk amounts, a request needs to be sent to the
central processing control for the request of blocking of some bandwidth by giving
the starting and ending time for the transfer of data. As the time slot is given, then
we can transfer any amount of data within that slot of time. The reassignment of
bandwidth and the multiple routing configurations are present and utilized in the
system. As per the results, the working of the system is excellent in the performance,
and good encouraging results were obtained from the system.
Keywords Advance reservations ·Bandwidth allocation ·Time slot ·Routing ·
Multiple routing ·Wired networks ·Wireless networks ·Bulk transfers
1 Introduction
The progress of communication networking together with the registering and capacity
advances is drastically changing the ways in recent days. Another term e-science has
S. NagaMallik Raj (B) · N. T. Rao
Department of Computer Science & Engineering, Vignan’s Institute of Information Technology
(A), Visakhapatnam, Andhra Pradesh, India
e-mail: mallikblue@gmail.com
S. Neeraja
Department of Computer Science & Software Engineering, Lendi Institute of Engineering and
Technology, Jonnada, Vizianagaram, Andhra Pradesh, India
D. Bhattacharyya
Department of Computer Science & Engineering, Koneru Lakshmaiah Education, Vaddeswa-
ram, Guntur, Andhra Pradesh 522502, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_6
61
62 S. NagaMallik Raj et al.
developed to portray the extensive scale science through worldwide joint efforts
empowered by systems. The expectation was reached by access to substantial scale
information accumulations, registering assets, and superior perception. All around
cited e-science cases incorporate high-vitality, high-energy nuclear physics, radio
stargazing, geosciences and atmosphere ponders. The requirement for transporting
expansive volume of information in e-science has very much contended [1–3].
To address the issue of e-science, the current paper contemplates admission
control, control transfer and booking calculations for high-transmission capacity
information into systems. The outcomes just propel the learning and procedures here
in addition to compliment the convention, engineering, and framework. The currently
considered admission control, control transfer and planning calculations handle two
classes of occupations, mass information exchange and a base data transfer capacity.
To transfer bulk amount of data through the networks, we need to concentrate
on two points. They are the speed of data in time and the space of the data to be
transferred in networks. They can be discussed in the following section as follows,
1. On-request mode: The data is to be transferred from source to the destination
whenever there is a request from the users to transfer bulk amount of data. The
data transfer will start from the request mode working.
2. In-advance mode: If the data needs to be transferred in bulk amounts, this option
will be chosen such that to transfer data. If we know the transfer of data, we need
to block the slot for transferring of data, to access such facility we need to utilize
the node of advance mode. We need to make the reservation of the slots for
transfer of data from source to destination.
2 Related and Proposed Work
The existing framework allows the confirmation of new demands as well as band-
width reallocation to existing work while not abusing the end-time prerequisites of
the present line of work. In the first place, it writes off the present stream job to the
old work on the future time cuts and improves the system to its special limit. The data
transmission of existing professions could be reallocated in the solitary link instance
nonetheless not in the system instance. The courses as well as move rate of existing
professions are unchanged [4, 4].
In the proposed framework, admission control, control transfer and booking calcu-
lations handle two classes of employment for mass information exchange etc. A mass
exchange isn’t touchy to the system delay yet might be delicate to the conveyance
due date. It is helpful for disseminating high volumes of logical information, which
right now regularly depends on ground transportation of the capacity media. The
MBG class is helpful for constant rendering or perception of information remotely.
In our system, the calculations for taking care of mass exchange additionally contain
the fundamental elements of those for taking care of the MBG class. Hence, we
will just spotlight on the mass exchange. One recognizing highlight in this inves-
tigation is that each activity demand can be made ahead of time. If an occupation
Earlier Selection of Routes for Data Transfer In Both … 63
is conceded, as controlled by the admission control calculation, the system ensures
that it will complete the information exchange for the activity before the asked for
the end time. The test is how to give this assurance while keeping up the proficient
use of the system assets and keeping the demand dismissal proportion low. The
way toward deciding the way of information exchange is known as planning [6].
The outcome is enormously enhanced productivity to arranging asset use. Creating
comparable conventions and adding new segments to the current toolbox in the help
of our calculations are among the future assignments [7].
In this paper, admission control, control transfer, and planning calculations for
high transmission capacity information moves in explore systems. The outcomes
won’t just propel the information and methods here yet in addition compliment
the convention, engineering, and foundation extends. Currently, the progress was
observed in the help of e-science and matrix processing by giving more produc-
tive system asset reservation and administration calculations. Current admission
control and booking calculations handle two classes of employment, mass informa-
tion exchange and those that require a base transfer speed ensure. The huge exchange
of data isn’t touchy to the system delay however might be delicate to the conveyance
due date. It is valuable for circulating high volumes of logical information, which at
present frequently depends on ground transportation of the capacity media. The way
toward deciding the way of information exchange is known as scheduling [4, 8–11].
3 Modules Description
The main modules in the current article are as follows,
Functional Requirements
3.1 Path Reservation
In this component, the customer’s attempts to hold way to course booking subtle
aspects start time, end-time, day, and method. Prior to holding way, the admission
control and control transfer checks ease of access for determined method officially
held or not. Booking plot for mass exchange, which checks book red profession time
with asked for start time as well as end time for the task and tries to discover a way
can match the entire work on that meantime. In reservation frame, it will certainly
collect the begin time, end time, date, goal, source to look for ease of access. At that
point in establishing the moment cut with bandwidth and document price quote. Else
it will certainly ask for that the client provides the right time term [12, 13].
64 S. NagaMallik Raj et al.
3.2 Minimum Bandwidth Allocation
Admission control, control transfer, and planning calculations handle two classes
of employment; mass information exchange and assign least data transmission to
ensure for a specific time in a specific way. So, information can be moved within the
opportunity to the goal. The transmission capacity can’t be utilized by others. The
requirement for effective system asset usage is particularly significant with regards
to bookings ahead of time and expansive record sizes or dependable streams. As
contended at there is an unfortunate wonder known as transfer speed fracture. The
least complex case of data transfer capacity discontinuity happens when the interim
between the end time of one employee and the start of another activity isn’t suffi-
ciently long for some other activity asks. A mass exchange demand may alternatively
indicate a base data transfer capacity as well as a most extreme transmission capacity.
Significantly, more parameters can be included, if necessary, like an expected range
for the requested measure or for the last days when the exact data is obscure.
3.3 Time Management
Employment’s point of view, it is alluring to have shorter reaction time. Each activity
demand can be made ahead of time and can determine a begin time and an end time.
The data transmission allocated to a specific way of an occupation stays steady for
the whole time cut, however, it might change the exchange time. That implies the
exchange might be finished inside the time and remaining time reservation will be
assigned for (open clients).
Rejection proportion: This is the proportion of the number of occupations
rejected and adds up several employment demands. From the system’s viewpoint, it
is alluring to concede; however, many employments as could be expected under the
circumstances. From the client’s viewpoint, it is alluring to have a little possibility
of employment dismissal [14].
3.4 Admission Control
To attend to the issue of e-science, this paper thinks of admission control as well
as planning estimations for high-transmission capacity information exchanges in
check into systems. Confirmation controller is brought together different which look
after work bookings as well as transfers. It gathers reservation asks for from hub
and check for accessibility then it rejects or recognizes the activity. It enhances the
system possession usage or reduced the activity dismissal percentage; the system
controller cares for advancement problems in settling on admission control as well
as booking options [15, 16].
Earlier Selection of Routes for Data Transfer In Both … 65
Fig. 1 The default graph existing the buttons send and reserve
Non-Functional Requirements
Adaptability: It is an appealing building of a structure, a system, or a procedure,
which shows its ability to either handle creating actions of work in a stylish way or to
be immediately broadened. A uniform of relevance is inferred, where the versatility
of a company infers that the fundamental strategy uses the potential for monetary
growth inside the organization.
The architecture model of the current system considered was explained in the
Figure 1 in detail and the architecture model of the designed system was as follows.
The admission control mechanism and the control transfer processes used to make
bulk transfer of data in both wired and wireless networks was shown in Figure 2
(Table 1).
4 Test Cases
The current system was tested for various cases such that to analyze the performance
of the current proposed system. The system was tested for almost 10 cases, the results
were displayed in the form of the tabular values, and the results were discussed in
detail in the above shown table.
5 Conclusion
The current proposed techniques go for adding to the administration and route allot-
ment of research systems for information concentrated e-science. The requirement
66 S. NagaMallik Raj et al.
Fig. 2 The admission control and control transfer show all the reserved paths in various times in
the database
for substantial record exchange and high-transmission capacity, low-inactivity orga-
nize ways is discussed in detail in the paper. These selected lists of routes can be
used and identified for sending and receiving the emergency data that can be used
for future purpose and other in near future applications and also used for other
set of applications in the mode of service to the society and to the public in the
society. The open doors lie in the way that examination systems are substantially
littler in estimate than people in general. This task consolidates the accompanying
novel components into a strong system of admission control, control transfer and
stream booking. Early bookings for mass exchange and least transfer speed ensured
movement, multipath directing, and data transmission reassignment by means of
occasional re-enhancement. By booking or blocking some set of routes that will help
us sending the emergency data to the end users whenever there is a need of sending
the emergency data. Hence, in this regard the selection and identification of the freely
available slots and less utilized routes were available in the total list of routes. They
are also available for the data or packets of data to be transmitted from sender to
receiver. To deal with begin and end time prerequisite of bookings ahead of time,
it distinguishes a reasonable group of discrete time-cut structures, to be specific the
consistent cut structures.
6 Future Enhancement
In our system we utilized streamlined control (admission control as well as control
transfer) to reserve the course to ensure that in future container neck problem might
Earlier Selection of Routes for Data Transfer In Both … 67
Table 1 Various test cases
Module Test scenarios Test case ID Test case
description
Test data Steps Expected results Results
Reservation on Verify the buttons
and text- boxes for
reservation
1Ensure that the
username and
password
Advance
application should
be available
1. Open the
application
2. Verify the
buttons, text boxes
are available
Login form should
be successfully
Login form was
successful
Reservation on Verify the graph of
the system
2Ensure that the
creationof
graph of the
system
Advance
application should
be available
1. Open the
application
2. Verify that the
application is
navigated to the
next page
Application should
be navigated to the
next page
Application on
navigated to the
next page
Reservation on Verify the check
list node
3Ensure that the list
node for the
destination
Advance
application should
be available
1. Open the
application
2. Verify that the
list box of the
destination node is
available or not
Verified list box of
the destination on
node should be
available
Verified list box of
the destination node
(continued)
68 S. NagaMallik Raj et al.
Table 1 (continued)
Module Test scenarios Test case ID Test case
description
Test data Steps Expected results Results
Reservation on Verify the start
time, end time, and
date
4Ensure that the
start time, end
time for reaching
the messages to
the destination
Advance
application should
be available
1. Open the
application
2. Select the start
time, end time and
date.
3. Verify the start
time and end time,
date that are
selected at the time
of reservation
Verified the list
boxes for start time,
end time, date
should be available
Verified the list
boxes for start time,
end time, date
should be available
Reservation on Verify the check
availability
5Ensure that the
path is available or
not
Advance
application should
be available
1. Open the
application
2. Click on check
availability
3. Verify that the
path is available or
not
Verified that there
served path should
be available
Verified that the
reserved path should
be available
Reservation on Verify the reserve
button
6Ensure that the
click on reserve
button then
displays window
but reserve path is
successful or not
Advance
application should
be available
1. Open the
application
2. Click on reserve
button
3. Verify the
reserve path is
successful getting
or not
Verified the reserve Verified the reserve
Earlier Selection of Routes for Data Transfer In Both … 69
occur, to solve this trouble we can save the reservation data in appropriate router
(dis-tributed manner). Likewise, we can make use of a few other effective organizing
for- mulas for scheduling.
References
1. Satyanarayana KV, Rao NT, Bhattacharyya D, Hu (2022) Identifying the presence of bacteria
on digital images by using asymmetric distribution with k-means clustering algorithm.
Multidimension Syst Signal Process 33(2):301–326. https://doi.org/10.1007/s11045-021-008
00-0
2. Chandra Sekhar P, Thirupathi Rao N, Bhattacharyya D, Kim T (2021) Segmentation of natural
images with k-means and hierarchical algorithm based on mixture of pearson distributions. J
Sci Ind Res 80(8):707–715. Retrieved from www.scopus.com
3. Bhattacharyya D, Dinesh Reddy B, Kumari NMJ, Rao NT (2021) Comprehensive analysis on
comparison of machine learning and deep learning applications on cardiac arrest. J Med Pharm
Allied Sci 10(4):3125–3131. https://doi.org/10.22270/jmpas.V10I4.1395
4. Bhattacharyya D, Doppala BP, Thirupathi Rao N (2020) Prediction and forecasting of persistent
kidney problems using machine learning algorithms. Int J Current Res Rev 12(20):134–139.
https://doi.org/10.31782/IJCRR.2020.122031
5. Mandhala VN, Bhattacharyya D, Vamsi B, Thirupathi Rao N (2020) Object detection using
machine learning for visually impaired people. Int J Current Res Rev 12(20):157–167. https://
doi.org/10.31782/IJCRR.2020.122032
6. Bhattacharyya D, Kumari NMJ, Joshua ESN, Rao NT (2020) Advanced empirical studies on
group governance of the novel Corona virus, Mers, Sars and Ebola: a systematic study. Int J
Current Res Rev 12(18):35–41. https://doi.org/10.31782/IJCRR.2020.121828
7. Asish Vardhan K, Thirupathi Rao N, Naga Mallik Raj S, Sudeepthi G, Divya Bhattacharyya
D, Kim T (2019) Health advisory system using IoT technology. Int J Recent Technol Eng
7(6):183–187. Retrieved from www.scopus.com
8. Eali SNJ, Bhattacharyya D, Nakka TR, Hong S (2022) A novel approach in bio-medical image
segmentation for analyzing brain cancer images with U-NET semantic segmentation and TPLD
models using SVM. Traitement Du Signal 39(2):419–430. https://doi.org/10.18280/ts.390203
9. Doppala BP, NagaMallik Raj S, Stephen Neal Joshua E, Thirupathi Rao N (2021) Automatic
determination of harassment in social network using machine learning. https://doi.org/10.1007/
978-981-16-1773-7_20 Retrieved from www.scopus.com
10. Eali SNJ, Rao NT, Swathi K, Satyanarayana KV, Bhattacharyya D, Kim T (2018) Simulated
studies on the performance of intelligent transportation system using vehicular networks. Int J
Grid Distrib Comput 11(4):27–36. https://doi.org/10.14257/ijgdc.2018.11.4.03
11. Joshua ESN, Battacharyya D, Doppala BP, Chakkravarthy M (2022) Extensive statistical anal-
ysis on novel coronavirus: towards worldwide health using apache spark. https://doi.org/10.
1007/978-3-030-72752-9_8. Retrieved from www.scopus.com
12. Joshua ESN, Bhattacharyya D, Chakkravarthy M (2021) Lung nodule semantic segmentation
with bi-direction features using U-INET. J Med Pharm Allied Sci 10(5):3494–3499. https://
doi.org/10.22270/jmpas.V10I5.1454
13. Joshua ESN, Bhattacharyya D, Chakkravarthy M, Kim H, -. (2021) Lung cancer classification
using squeeze and excitation convolutional neural networks with grad cam++ class activation
function. Traitement Du Signal 38(4):1103–1112. https://doi.org/10.18280/ts.380421
14. Joshua ESN, Chakkravarthy M, Bhattacharyya D (2021) Lung cancer detection using impro-
vised grad-cam++ with 3D CNN class activation. https://doi.org/10.1007/978-981-16-177
3-7_5. Retrieved from www.scopus.com
70 S. NagaMallik Raj et al.
15. Neal Joshua ES, Bhattacharyya D, Chakkravarthy M, Byun Y (2021) 3D CNN with visual
insights for early detection of lung cancer using gradient-weighted class activation. J Healthc
Eng. https://doi.org/10.1155/2021/6695518
16. Neal Joshua ES, Chakkravarthy M, Bhattacharyya D (2020) An extensive review on lung
cancer detection using machine learning techniques: a systematic study. Revue d’Intelligence
Artificielle 34(3):351–359. https://doi.org/10.18280/ria.340314
Identifying River Drainage
Characteristics by Deep Neural Network
Vithya Ganesan, Tejaswi Talluru, Manoj Challapalli, and Chandana Seelam
Abstract This work provides environmental protection and sustainable develop-
ment to manage network of reservoirs and canals for identifying inner water link
under river. Continuous monitoring of the river width, speed, flow, and longitude
images of the river are analyzed by time series and AIoT technique to predict the
path and trace the direction of inner and outer flow of river. At the same time, get
prediction of data and images on soil alleviation and erosion. Extract the significance
of rivers / drainage images from high-resolution multispectral satellite by framework
is developed to identify river drainage characteristics such as inner water link, predic-
tion of river path by its width and longitude and compare the images after natural
calamities. Analyzed the multi spectral images to develop digital elevation maps of
river drainage features and provide guidance for disaster preparedness.
Keywords River feature extraction ·River topology ·Deep learning ·River
skeleton ·River models
1 Introduction
To define and test various drainage images by using multispectral images is real
challenge in deep learning. Analyze the multispectral images for identifying river
drainage width, depth, angle of elevation by identify the characteristics of river
drainage based on its capacity, latitude, longitude, and volume by using multispec-
tral river image. The proposed model, Deep Image analysis is employed to generate
drainage map with its volume and identifying the drainage deviations before and after
the natural calamity. Motivation and objective of this work is by deploying a frame-
work to extract feature of rivers drainage images from high resolution multispectral
images to improve the study of evolution of drainage characteristics to identify nature,
structure of rocks, topography of river and land slope.
V. Ganesan · T. Tal l uru · M. Challapalli (B) · C. Seelam
CSE, Koneru Lakshmaiah Education Foundation, Guntur, Andhra Pradesh 522302, India
e-mail: manojchallapalli93@gmail.com
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_7
71
72 V. Ganesan et al.
Multispectral Images is the main source of information for generating and
updating topographic river and drainage data. The extraction of rivers being an appli-
cation that deserves considerable attention, especially in large areas of vegetation and
cultivation.
To identify the different drainage patterns from high-resolution images gives more
clarity to identify the skeleton of drainage system. On continuous monitoring of
drainages by its width, depth, and elevation angle ascertain with climatic condition
and natural calamities is required by new analysis tool. A suitable AI and deep
learning and will be helpful to analysis the drainage flow and path.
2 Related Work
Erosion and sediment deposition by rivers and streams paving remarkable passages
[1]. The world economy is directly linked with river to enhance the quality of life
[2]. Developing a hybrid intelligent framework for extracting river drainage images
to act as a road map for identifying regularities and irregularities, certainty, and
uncertainties in river/drainage flow path by hydrography image [3] and it is shown
in Figs. 1 and 2.
2.1 Identifying Various Size and Shape of River Drainage
River curves accelerate the erosion process [4] and river flow images by erosion are:
•Horseshoe: U-shaped water bodies due to erosion by river flow
•Scars: River sediment based on its speed, volume, and perineal. The sediment
deposition
Fig. 1 River/drainage flow path
Identifying River Drainage Characteristics by Deep Neural Network 73
Fig. 2 River/drainage hydrography
•Sandbar, side bar, and scroll bar: it is characterized by the deposition of sediment
and images of sandbar on the river path example: Krishna river side, sand, and
scroll bars.
•Cut bank, Relic Channels, and secondary channel are other types of sediment
scars
•Delta: river divides into several smaller parts form a triangular area
•Drainage patterns: Based on land elevation and rock types of the drainage pattern
varies by its geologic structures [5].
The river /drainage issues delve by the following phases, which helps to estimate
width and longitude of the river [6]. Consequently, from literature survey, an engi-
neering approaches is required to identify the river elevation, drainage characteristics
and river environment are examined by the visual localization and knowledge base
framework of the river [7]. It is segmented in to two phases.
They are 1. Identifying various size and shape of river drainage 2. Generate river
drainage image resolutions by spatial, spectral, angular, and temporal model. 3.
Elevate a drainage map for the above model to predict the drainage characteris-
tics. A knowledge-based framework invokes the following [8]. Identify high reso-
lution multispectral river drainage images, Classify River drainage path flow by
deep learning. Train by Knowledge base engineering to discern regularities and
irregularities in River drainage path and shape [9].
2.2 Generate River Drainage Image by Spatial, Spectral,
Angular, and Temporal Model
The appearance of the selected images is enhanced by applying spatial, spectral,
angular, and temporal model using polyline layers technique to sharpen the edges of
features in river drainage images. Identify and analysis various river images by deep
74 V. Ganesan et al.
Fig. 3 River drainage image analysis by deep learning model
learning model such as spatial, spectral, angular, and temporal model are shown in
Fig. 3.
3 Result
Multispectral river drainage image: Uses shape files and pyshedes to identify
longitude, latitude, and elevation of the river flow, and the result is s hown in Fig. 4.
Using deep learning, extract the various size and shape of river drainage and its
resultant image is shown in Fig. 5.
By using elevation data set, river flow direction for spatial model of river drainage
image with flow direction grid is generated and it is shown in Fig. 6.
Fig. 4 Digital elevation map for river flow
Identifying River Drainage Characteristics by Deep Neural Network 75
Fig. 5 Drainage network
Fig. 6 River flow direction for spatial model of river image
River flow accumulation for spectral model of the image with flow accumulation
is generated by using shape and TIFF files and it is shown in Fig. 7.
4 Discussion
In angular model, the total river flow distance, depth and its flow distance in are
visualized by using pyshede, and it is shown in Fig. 8.
In temporal model, river phenomenon such as percent impervious area is calcu-
lated by latitude and longitude on x and y-axis to identify the temporal data about
water images, and it is shown in Fig. 9.
76 V. Ganesan et al.
Fig. 7 River flow accumulation for Spectral Model of river image
Fig. 8 River flow direction in angular model
Soil texture analysis is required to identify t he longitude and latitude of the
riverbed. To track the river flow path and deviations, texture analysis is helpful.
Figure 10 shows the different colors of soil texture type, and Fig. 11 shows raster
image of soil texture analysis for diagnosing riverbed. From River flow direction for
spatial model of river image, River flow accumulation for Spectral model of river
image and River flow direction in angular model shows the river phenomenon such
as length, breadth, and elevation. Temporal model of river phenomenon, Soil texture
analysis for riverbed identification and Raster image of Soil texture is used to ascer-
tain delineated catchment by comparing the images periodically. In Fig. 12 shows
delineated catchment of river flow direction is visualized from sample data set of
river elevation. From the periodical comparison, if any anomaly is detected, then it
is identified that minor calamity is occurred near the riverbed.
Identifying River Drainage Characteristics by Deep Neural Network 77
Fig. 9 Temporal model of river phenomenon
Fig. 10 Soil texture analysis for riverbed identification
5 Conclusion
Morphological assessment of the river is necessary for environmental planning which
is used to understand the current situation and its possible route changes. Normally,
extraction of river images is by preprocessing and analysis by filtering, smoothing,
delineation technique. In addition, improving multidimensional river images by aver-
aging a sequence of images, with its depth, latitude and longitude is implemented
by different models to predict flow and direction of the river. It is used to generate
drainage map with its volume and identifying the drainage deviations before and
after the natural calamity.
78 V. Ganesan et al.
Fig. 11 Raster image of Soil texture
Fig. 12 Delineated catchment after the minor calamity
References
1. Rao YKN, Saito K, Ch V, Kumar S, Kubo S, Pandey Z, Li G, Demudu AS. Rajawat: holocene
evolution and Anthropocene destruction of the Krishna Delta on the east coast of India: Delta
lobe shifts, human impacts, and sea-level history. Marine Geol (106229)
2. Precision mapping of boundaries of flood plain river basins using high-resolution satellite
imagery: a case study of the Varuna river basin in. J Earth Syst Sci 128(4):105–105
Identifying River Drainage Characteristics by Deep Neural Network 79
3. Hansda S, Das VK, Debnath K (2022) Temporal modulation of turbulence structure over
progressive erosion boundary under influence of wave current combined flow. Environ Fluid
Mech
4. Li M, Wu B, Chen Y, Li D (2022) Quantification of river network types based on hierarchical
structures. CATENA 211
5. Arévalo OJ, Colombera L, Mountney NP, Basilici G, Marcus VT, Soares: variations in water
discharge at different temporal scales in a mud-prone alluvial succession: the Paleocene-Eocene
of the Tremp-Graus Basin. Spain. https://doi.org/10.1016/j.sedgeo. 2022.106122
6. Reddy RP, Srija K, Karthi SS, Geetha RP (2020) Evaluation of water body extraction from
satellite images using open-source tools. In: Intelligent systems, technologies and applications.
Advances in intelligent systems and computing. vol 910. Springer
7. Bajirao TS, Kumar P (2021) Geospatial technology for prioritization of Koyna River basin of
India based on soil erosion rates using different approaches. Environ Sci Pollut 28:35242–35265
8. Wood1* DJ, Brown1 CRM, Doyle1 L, Smith1 H, Waller1 S, Jba1 EFW. Identification of river
defences from digital terrain models using deep learning. Risk Management
9. https://unesdoc.unesco.org/ark:/48223/pf0000372985.locale=en
A Review on Optimal Deep Learning
Based Prediction Model for Multi Disease
Prediction
Aneel Kumar Minda and Vithya Ganesan
Abstract Healthcare data collecting and processing is one of the most worrisome
troublesome to optimize the methodology. With the advent of the digital era and
technological advancements, a vast quantity of multidimensional data on patients
is created, including clinical factors, hospital resources, illness diagnostic informa-
tion, patients’ records, and medical equipment. The enormous, dense, and complex
data must be processed and evaluated to extract knowledge for effective decision-
making. Medical data mining offers a lot of potential for uncovering hidden patterns
in medical data sets. By identifying significant patterns and detecting correlations and
relationships among many variables in huge databases, the use of various data mining
tools and machine learning approaches has changed healthcare organizations. This
review paper identifies an importation in the medical data, providing and comparing
existing data for the future course of action.
Keywords Health care ·Health prediction ·Disease predication ·Data science for
healthcare ·Medical care data analysis
1 Introduction
Technology combines multiple analytic methodologies with modern and complex
algorithms, allowing for the exploration of massive amounts of data [1–4]. It is used
in healthcare to gather, organize, and analyze patient data in a systematic manner.
It may be used to identify inherent inefficiencies and best practices for providing
better services, which may lead to improved diagnosis, better medicine, and more
successful treatment, as well as a platform for a deeper knowledge of the mechanisms
in practically all elements of the medical domain [5–10]. Overall, it assists in the
A. K. Minda (B)
International SOS, Dubai, United Arab Emirates
e-mail: mak.msbi5@gmail.com
V. Ganesan
CSE, Koneru Lakshmiah Education Foundation, Guntur, India
e-mail: vithyaganesan@kluniversity.in
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_8
81
82 A. K. Minda and V. Ganesan
early detection and prevention of disease epidemics by searching medical databases
for pertinent information. The process of determining a condition based on a person’s
symptoms and indicators is known as medical diagnosis. In the diagnostic process,
one or more diagnostic procedures, such as diagnostic tests, are performed [11–13].
Diagnosis of chronic illnesses is a vital issue in the medical industry.
Since it is based on many symptoms. It is a complex procedure that frequently
leads to incorrect assumptions. When diagnosing illnesses, the clinical judgment is
based mostly on the patient’s symptoms as well as the physicians’ knowledge and
experience [14–19]. Furthermore, when medical systems evolve and new treatments
become available, it becomes more difficult for physicians and doctors to stay up
with the current innovations in clinical practice. For effective therapy, medical prac-
titioners and doctors must be well-versed in all pertinent diagnostic criteria, patient
history, and a mix of medication therapy. However, mistakes are possible since they
make judgments instinctively based on information and experience gained from past
experience with patients. Because of factors such as multitasking, restricted analysis,
and memory capacity, their cognitive capacities are restricted [20, 21]. As a result, it
is difficult for a physician to make the right judgment on a consistent basis if he is not
supported by clinical tests and patient history i nformation. Even experienced physi-
cians can benefit from a computer-aided diagnostic system in making sound medical
judgments [22–25]. Thus, medical professionals are very interested in automating the
diagnosis process by integrating machine learning techniques with physician exper-
tise. Data mining and machine learning approaches are making significant efforts to
intelligently translate accessible data into valuable information in order to improve
the diagnostic process’s efficiency. Several studies have been conducted to explore
the use of machine learning in terms of diagnostic abilities. It was discovered that,
when compared to the most experienced physician, who can diagnose with 79.97%
accuracy, machine learning algorithms could identify with 91.1% correctness [6, 26–
28, 30]. Machine learning techniques are explicitly used to illness datasets to extract
features for optimal illness diagnosis, prediction, prevention, and therapy.
2 Literature Review
In 2019, Usama et al. [29] have implemented a self-attention-based recurrent convolu-
tional neural network (RCNN) representation by “real-life clinical text data collected
from a hospital in Wuhan, China”. The proposed model had learned the elevated
semantic features automatically from medical text using an indirect association
surrounded by convolution. The clinical text also had limitations as a result the
RCNN capability with the self-attention method was examined. Hence, through the
self-attention model, convolve features were focused which had the efficient signif-
icance in the clinical text using the measurement of the probability between each
convolve feature by softmax. This model was estimated on the dataset and the metrics
as accuracy and recall were analyzed. The obtained outcomes have proved the better
A Review on Optimal Deep Learning Based Prediction Model … 83
accuracy of the proposed model than numerous existing approaches in detecting the
disease of cerebral infarction.
In 2019, Jiang et al. [30] have implemented a novel multi-task learning approach
for the prediction of the cognitive diseases and identification of the major predictive
biomarkers exactly based on a low rank constrained regularization and correlation-
aware sparse. The new algorithm of multi-task learning was proposed for the opti-
mization of non-smooth convex. The experiments were carried out based on the
Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset for evaluating the
optimization.
Mainly, the baseline magnetic resonance imaging (MRI) features were attained
depends on the predicted cognitive scores from the numerous time points. The results
were shown that the correctness and rationality of the proposed multi-task learning
method and the prediction of disease progression were reliable.
In 2020, Lei et al. [31] have implemented a framework for the prediction clinical
scores that were depending on longitudinal various data time points. This framework
involved in three stages like “feature selection based on chronotropy regularized
joint learning, feature encoding based on deep polynomial network, and ensemble
learning for regression via the support vector regression method”. For the prediction
of scores, two scenarios were planned such as scenario 1 and scenario 2. In scenario 1,
the prediction of the longitudinal scores was achieved for the baseline data, whereas
scenario 2 was used for obtaining the predicted scores using all the previous data time
points, which could develop the accuracy of the prediction scores. It is applied to
resolve the data incompleteness. By using the public database of ADNI, the developed
structure could efficiently show the association among experimental score and data
of MRI.
In 2019, Khan et al. [1] have applied the network analysis and data mining proce-
dure on the hospital data regarding admission and discharge for recognizing the
chronic patient’s disease or comorbidity tracking. Based on this, a chronic disease
risk prediction framework was developed and tested with “Australian healthcare
context dataset” for the prediction of Type 2 Diabetes (T2D) risks. Some of the risk
factors for the predictions are the clustering membership, comorbidities occurrences,
and transition patterns by procedures of social network based and numerous graph
theories. Moreover, the exploratory procedure was developed with three predictive
methods like parameter optimization, regression, and classification trees. All three
prediction approaches produced for the graph theory, which had the highest-ranking
depended on the “comorbidity prevalence” and “transition pattern match” scores.
Finally, the overall prediction accuracy was enhanced by utilizing administrative
data. In 2018, Hashem et al. [14] have compared and estimated various ML tech-
niques for the prediction of fibrosis in advanced stage using the combination of the
clinical information and serum biomarkers for the classification of system develop-
ment. For the advanced fibrosis risk prediction model various methods like particle
swarm optimization (PSO), DT, multilinear regression, and GA were developed.
This proposed model has performed better for producing inexpensive, numerical,
and accurate outcomes in real-time environment. This model was used for predicting
84 A. K. Minda and V. Ganesan
the advanced liver fibrosis among the correlation coefficient and has attained high
accuracy when compared to DT and PSO algorithms.
In 2019, Mohan et al. [2] have implemented an approach for f orecasting some
features using ML methods, which results in the accuracy improvement for the
cardiovascular disease prediction. The prediction model was developed by dissimilar
features combination and numerous classification approaches. The performance of
the proposed model has produced an enhanced accuracy level with the hybrid random
forest with a linear model (HRFLM), thus the accuracy and time of testing were
improved. The proposed model approved the data records into two classes in support
vector machine (SVM) and ANN for the additional analysis process. The back
propagation neural network (BPNN) by classification approach was implemented,
wherever the generation of the hypertension gene sequence occurred.
In 2019, Aliberti et al. [3] have discussed the crisis of automatic prediction of
glucose level for multi-patient data. By analyzing the training set of multipatient,
the glucose level of the prediction model was analyzed for predicting the upcoming
level of glucose for a new patient. Further, the major contribution of this system was
based on two processes: (1) discover the prediction model by a group of CMGS data
from a mixed group of diabetic patients, and this probably improved the ability of
generalization and reduced the over-fitting risks of the model; (2) plan and evaluate
the diverse category of prediction system, the prediction results were analyzed from
the analytical and the clinical point of view.
In 2019, Almansour et al. [4] have focused on various ML (Dahiwade et al. 2019)
and classification algorithms to “a dataset of 400 patients and 24 attributes” linked
with the chronic kidney disease (CKD) diagnosis. The classification methods like
ANN and SVM were used for experiments; then in the dataset, the entire missing
values were substituted using the mean value of the equivalent attributes. Subse-
quently, for the ANN and SVM, the optimized parameters were determined using
the alteration of the parameters. The proposed model was implemented using best
attained features and parameters. From the experimental results, the ANN presented
enhanced performance than SVM, with better accuracy.
3 Review on Traditional Multi-disease Prediction Models
In recent years, the disease diagnosis is based on the computer-aided model from
clinical data by DL approaches is an emerging and wide area of research. However,
there are some challenges exited in this field that have to be considered as the most
significant. A few of the major pros and cons are represented in Table 1. Based on
RNN in [29] in multi-disease prediction, it attains high prediction accuracy, and it
can handle any size of data, but the computational complexity is high due to recur-
rent nature, and training of RNN seems to be difficult when it operates on large
environment. Multi-task learning [30] is applied for Alzheimer’s disease prediction
provides consistent and efficient information is attained through multi-task learning.
It prevents overfitting. However, this method is hard to achieve for large experiments.
A Review on Optimal Deep Learning Based Prediction Model … 85
It can achieve only single modality data. For Alzheimer’s disease monitoring [31]
the deep polynomial network and ensemble learning is helpful even for unstructured
data, and this method can attain accurate prediction, still, the Problem formulation
is tough for large data sets. When the network becomes deeper, the optimization
problem may occur. By the data mining and network analysis techniques like predic-
tive training regression, tree classification, and parameter optimization [1] it obtains
good prediction accuracy and it promotes the quality of information. Yet, it faces
few difficulties while handling real-time datasets and more resources are required
for large administrative datasets. The ML is mainly efficient in attaining low robust
and computational time for overfitting problem. High accuracy is maintained. It iden-
tifies data easily. However, debugging of the problem is difficult in particle swarm
optimization and it has a low convergence rate. GA is computationally expensive. The
proposed HRFLM [2] for predicting cardiovascular disease produces the enhanced
prediction and accuracy level and improves the classification performance. While
compared to decision trees, it’s hard to compute, and is time-consuming algorithm.
CGMSs on considering feed forward networks (FNN) [3], and RNN. FNN does not
depend on input data, and the networks are able to generalize the entire network
by reading a few data sets. However, FNNs require lengthy training sessions, and
practical issues may arise in RNN. For the prediction of CKD [4], SVM effectively
performs better than the ANN, and both methods store information on the entire
network. Even though SVM is efficient, it is not suitable for large applications. Its
hardware dependence affects the performance. These drawbacks should be consid-
ered to motivate the upcoming researchers in developing more advanced techniques
for predicting multi-disease in health care.
Objectives.
To extract the most relevant features training the deep learning classifiers.
To select the most relevant features for higher reliability and lower computational
complexity.
To design a hybrid deep learning framework for precise disease prediction.
To fine-tune the hyper parameters of the deep learning model with optimization
techniques for enhancing the prediction performance.
To introduce a new self-improved optimization model for enhancing the conver-
gence of the solutions, thereby solving the optimization problem.
4 Methodology
Prediction of diseases is a major factor for medical organizations toward creating the
best medical decisions. In medical treatment, wrong decisions may result in treat-
ment delay or even death. Various disease prediction models are studied and many
limitations are affecting the treatment. The major challenges in medical sector are
dissimilar aggregations, in which the data sources may be asynchronous nature into
significant indicators of personal health. Earlier, professionals of healthcare faced the
86 A. K. Minda and V. Ganesan
Table 1 Methodology and Challenges
Author Methodology Features Challenges
Khan et al. [1] RNN Can attain high
prediction accuracy
RNN can handle any
size of data
Computational
complexity and training
are high due to
recurrent nature
Mohan et al. [2] Multi-task learning The consistent and
efficient information is
attained to multitask
learning. It large
prevents overfitting
This method is hard
achieve for the through
experiments
It couldn’t perform on
multi-modality data
Aliberti et al. [3]Deep polynomial
network, and ensemble
learning
It obtains more
accurate predictions
It is very useful when
thedataisnot in
particular structure
Problem formulation is
tough for large datasets
when the network
becomes deep
Almansour et al. [4] Predictive training
regression,
classification, and
parameter optimization
It obtains good
prediction tree
accuracy and quality
information
Difficult to handle
real-time datasets More
resources are required
for large administrative
datasets
Zhao et al. [5]PSO, DT,
multi-regression, and
models
It is efficient in
attaining low linear
computational time
and robust to
overfitting.
High accuracy is
maintained. It
identifies data easily
Debugging of the
problem is difficult. It
has low convergence
rate
GA is computationally
expensive
Wang et al. [6] HRFLM It produces the
enhanced prediction
and accuracy level
Compared to decision
trees, it’s hard to
compute
Brand et al. [7]FNNs and RNN It does not impose on
the input data
The networks can
generalize issue may
reading few data sets
FNNs require lengthy
training the entire
network by practical
arise in RNN
Kumar et al. [8]SVM It stores information
on the entire network.
SVM achieves more
accuracy than ANN
SVM is not suitable for
large applications
It’s hardware
dependence affects the
performance
challenges to gather and estimate the enormous amount of data for successful treat-
ments and predictions because of fewer tools or technologies. For disease prediction
based on existing approaches, f ewer variables are considered “such as age, weight,
height, gender, and more”. In contrast, the ML approach uses more variables, which is
A Review on Optimal Deep Learning Based Prediction Model … 87
based on computing devices. Thus, ML for disease prediction can attain better accu-
racy in the healthcare field. Prediction of future medical status is done by various
algorithms.
These algorithms help to construct models for data analyzing and delivery of
results, using the historical data and real-time data. By using ML, healthcare profes-
sionals decide improved assessment on diagnoses of patient’s data and treatment
choices, which lead to improving the services of healthcare. DL is the new and
important progression of ML, which is applied for efficient extraction of impor-
tant features from complex and huge datasets by using the hierarchical and stacked
learning approaches. DL could provide improved performance in numerous sectors
like recognition of speech, natural language processing, and recognition of images.
In this research work, a novel disease prediction model will be developed by
following four major phases: (1) data normalization (2) feature extraction, (3) feature
selection, and (4) prediction.
The proposed architecture for a multidisease-prediction framework using a hybrid
DL method. Various datasets are collected from benchmark datasets for conducting
the experiment are diabetes, hepatitis, lung cancer, liver tumor, breast cancer, COVID-
19, heart disease, Parkinson’s disease, Alzheimer’s disease, and its processing is
shown in Fig. 1.
In the feature extraction phase, the most relevant features like statistical features
(mean, median, and standard deviation as well), modified correlation, modified
skewness, modified entropy, and technical indicators-based features will be extracted.
First, the collected dataset values will be applied to the pre-processing phase,
where the normalization of values will be done here within the range of 0–1. Normal-
ization of data is applied for systematizing the non-structured data into structured
data. Data normalization is efficient for minimizing the redundancy of data and data
complication and also data integrity is enhanced. Further, the normalized attributes
are employed for the feature extraction.
From the extracted features, the reliable features will be selected with improved
chi-square model. The disease prediction model will be designed with hybrid deep
learning algorithms with optimized Bi-GRU and quantumNet, respectively. The opti-
mized Bi-GRU and quantumNet are trained with the appropriate features acquired
from improved chi-square model.
To enhance the prediction accuracy of the projected model, the weight function
of Bi-GRU will be fine-tuned with new self-improved Honey Badger Algorithm
(SI-HBA) model. This SI-HBA model will be the conceptual enhancement of the
standard HBA [32] model. The HBA is inspired from the intelligent foraging behavior
of honey badger, to mathematically develop an efficient search strategy for solving
optimization problems. The dynamic search behavior of honey badger with digging
and honey finding approaches are formulated into exploration and exploitation phases
in HBA. Moreover, with controlled randomization techniques, HBA maintains ample
population diversity even toward the end of the search process.
The optimized Bi-GRU and quantumNet will run in parallel and independently.
The average of the final predicted outcome from optimized Bi-GRU and quantumNet
is computed, and it is the ultimate decision regarding the presence/absence of disease.
88 A. K. Minda and V. Ganesan
Fig. 1 Overview of disease prediction
5 Conclusion
The proposed model will be carried out in Python and the experimented outcome will
be investigated. The performance of the proposed model will be compared over other
state-of-the-models in terms of Type I and Type II measures. Here, Type I measures
are positive measures like accuracy, sensitivity, specificity, precision, negative predic-
tive value (NPV), F1Score, and Mathews correlation coefficient (MCC), and Type
II measures are negative measures like false positive rate (FPR), false negative rate
(FNR), and false discovery rate (FDR).
References
1. Khan A, Uddin S, Srinivasan U (2019) Chronic disease prediction using administrative data
and graph theory: the case of type 2 diabetes. Expert Syst Appl 136:230–241
2. Mohan S, Thirumalai C, Srivastava G (2019) Effective heart disease prediction using hybrid
machine learning techniques. IEEE Access 7:81542–81554
A Review on Optimal Deep Learning Based Prediction Model … 89
3. Aliberti A, Pupillo I, Terna S, Macii E, Cataldo DS, Acquaviva PEA (2019): A multi-patient
data-driven approach to blood glucose prediction. IEEE Access 7:69311–69325
4. Almansour NA, Syed HF, Khayat NR, Altheeb RK, Juri RE, Alhiyafi J, Alrashed S, Olatunji
SO (2019) Neural network and support vector machine for the prediction of chronic kidney
disease: a comparative study. Comput Biol Med 109:101–111
5. Zhao Y, Ma B, Jiang P, Zeng D, Wang X, Li S (2021) Prediction of Alzheimer’s disease
progression with multi-information generative adversarial network. IEEE J Biomed Health Inf
25(3):711–719
6. Wang T, Tian Y, Qiu RG (2020) Long short-term memory recurrent neural networks for multiple
diseases risk prediction by leveraging longitudinal medical records. IEEE J Biomed Health Inf
24(8):2337–2346
7. Brand L, Nichols K, Wang H, Shen L, Huang H (2020) Joint multi-modal longitudinal
regression and classification for Alzheimer’s disease prediction. IEEE Trans Med Imaging
39(6):1845–1855
8. Kumar M, Kumar A, Palaparthy VS (2021) Soil sensors-based prediction system for plant
diseases using exploratory data analysis and machine learning. IEEE Sens J 21(16):17455–
17468
9. Wu C, Gao R, Zhang Y (2019) mHMDA: human microbe-disease association prediction by
matrix completion and multi-source information. IEEE Access 7:106687–106693
10. Wang H, Huang Z, Zhang D, Arief J, Lyu T, Tian J (2020) Integrating co-clustering and
interpretable machine learning for the prediction of intravenous immunoglobulin resistance in
Kawasaki disease. IEEE Access 8:97064–97071
11. Dong Y, Sun Y, Qin C, Zhu W (2020) EPMDA: edge perturbation based method for miRNA-
disease association prediction. IEEE/ACM Trans Comput Biol Bioinform 17(6):2170–2175
12. Ge R, Zhang R, Wang P (2020) Prediction of chronic diseases with multi-label neural network.
IEEE Access 8:138210–138216
13. Liu P, Luo J, Chen X. miRCom: tensor completion integrating multi-view information to deduce
the potential disease-related miRNA-miRNA pairs. IEEE/ACM Trans Comput Biol Bioinform
14. Hashem S, Esmat G, Elakel W, Habashy S, Raouf SA, Elhefnawi M, Eladawy M, Elhefnawi M
(2018) Comparison of machine learning approaches for prediction of advanced liver fibrosis
in chronic hepatitis C patients. IEEE/ACM Trans Comput Biol Bioinform 15(3):861–868
15. Zhang J (2021) Multi-resemblance multi-target low-rank coding for prediction of cognitive
decline with longitudinal brain images. IEEE Trans Med Imaging 40(8):2030–2041
16. Zhang, Y, Ye F, Gao X. MCA-net: multi-feature coding and attention convolutional neural
network for predicting lncRNA-disease association. IEEE/ACM Trans Comput Biol Bioinform
17. Jiang P, Wang X, Li Q, Jin L, Li S (2019) Correlation-aware sparse and low- rank constrained
multi-task learning for longitudinal analysis of Alzheimer’s disease. IEEE J Biomed Health
Inform 23(4):1450–1456
18. Yang K. PDGNet: predicting disease genes using a deep neural network with multi-view
features. IEEE/ACM Trans Comput Biol Bioinform
19. Xuan P, Gao L, Sheng N, Zhang T, Nakaguchi T (2021) Graph convolutional autoencoder
and fully-connected autoencoder with attention mechanism based method for predicting drug-
disease associations. IEEE J Biomed Health Inform 25(5):1793–1804
20. Zhang Z, Ding J, Xu J, Tang J, Guo F (2021) Multi-scale time-series kernel-based learning
method for brain disease diagnosis. IEEE J Biomed Health Inform 25(1):209–217
21. Xuan P, Zhan L, Cui H, Zhang T, Nakaguchi T, Zhang W. Graph triple-attention network for
disease-related LncRNA prediction. IEEE J Biomed Health Inform
22. Peng W, Liu M, Dai W, Chen T, Fu Y, Pan Y. Multi-view feature aggregation for predicting
microbe-disease association. IEEE/ACM Trans
23. Dayun L, Junyi L, Yi L, Qihua H, Deng L. MGATMDA: predicting microbe-disease
associations via multi-component graph attention network. IEEE/ACM Trans Comput Biol
Bioinf
24. Song Y, Cui H, Zhang T, Yang T, Li X, Xuan P. Prediction of drug-related diseases through
integrating pairwise attributes and neighbor topological structures. IEEE/ACM Trans Comput
Biol Bioinforma
90 A. K. Minda and V. Ganesan
25. Prince J, Andreotti F, De Vos M (2019) Multi-source ensemble learning for the remote predic-
tion of Parkinson’s disease in the presence of source-wise missing data. IEEE Trans Biomed
Eng 66(5):1402–1411
26. Wu QW, Cao RF, Xia J, Ni JC, Zheng CH, Su Y. Extra trees method for predicting LncRNA-
disease association based on multi-layer graph embedding aggregation. IEEE/ACM Trans
Comput Biol Bioinform
27. Han G, Kuang Z, Deng L. MSCNE: Predict miRNA-disease associations using neural network
based on multi-source biological information. IEEE/ACM Trans Comput Biol Bioinform
28. Shi J (2019) Cascaded Multi-column RVFL+ classifier for single-modal neuroimaging-based
diagnosis of Parkinson’s disease. IEEE Trans Biomed Eng 66(8):2362–2371
29. Usama M, Ahmad B, Xiao W, Hossain MS, Muhammad G (2019) Self-attention based recurrent
convolutional neural network for disease prediction using healthcare data. Comput Methods
Programs Biomed 190:11–11
30. Jiang P, Wang X, Li Q, Li JLS (2019) Correlation-aware sparse and low-rank constrained
multi-task learning for longitudinal analysis of Alzheimer’s Disease. IEEE J Biomed Health
Inform 23(4):1450–1456
31. Lei B, Yang M, Yang P, Zhou F, Hou W, Zou W, Li X, Wang T, Xiao X, Wang S (2020) Deep
and joint learning of longitudinal data for Alzheimer’s disease prediction. Pattern Recognit
102:107247–107247
32. Fatma A, Hashim, Essam H, Houssein K, Hussain MS, Mabrouk W, Al-Atabany. Honey Badger
algorithm: new metaheuristic algorithm for solving optimization problems. Math Comput
Simulation 192:2022–2022
A Hybrid Multi-user Based Data
Replication and Access Control
Mechanism for Cloud Data Security
V. Devi Satya Sri and Srikanth Vemuru
Abstract Data replication-based cloud data access control mechanism plays a major
role in real-time cloud computing environment due to high computational cost and
memory. Hybrid data replication models play a vital role in the cloud-based applica-
tions for data recovery and security. Machine learning tools and techniques play an
essential role in the medical field and cloud computing applications. Most of the tradi-
tional machine learning models use static data partitioning and replication methods in
order to recover patterns from multiple virtual machines in cloud computing environ-
ment. In this work, a hybrid data replication and multi-user data access mechanism is
developed to provide strong data recovery and security in cloud computing environ-
ment. In this work, a machine learning based patterns are used for data partitioning
and security in the cloud server for cloud data security. Experimental results show
that the hybrid data replication model has better data replication time and storage
space than the conventional data replication models in cloud computing environment.
Keywords Data replication ·Data partitioning ·Cloud computing ·Data access ·
Medical patterns
1 Introduction
The access speed of the data should be increased to keep the load in the system
balanced. The two main factors for improving cloud performance are the scalability
and availability. The replication creates multiple copies of an existing entity [1]. The
creation of replicas is one of the important approaches for achieving this. Replica-
tion enhances resource availability. Replication also offers minimum access costs,
shared bandwidth usage and time delays through data Replication. In case of system
failure, the value of replication is transparent, flawless access to resources. Repli-
cation across a computer network can be extended to allow storage devices to be
V. D ev i S a t y a S r i · S. Vemuru (B)
Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation,
Vaddeswaram, Guntur District, A.P, India
e-mail: vsrikanth@kluniversity.in
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_9
91
92 V. Devi Satya Sri and S. Vemuru
located in physically separated installations. In the event of failure to ensure data
transmission, users access nearby replicas and increase the throughput. The data is
saved at more than one site with advantages. A system can operate using replicated
data when a server fails with the required data. This concept remains accessible.
Data are saved at several locations. The requested data is collected from the source
of the request. This increases the system’s performance. Replication benefits are
not provided without overheads of replicas being created, maintained, and updated.
Replication can improve performance significantly [2]. Replication technology takes
time to recover data from other sites and restart service again. There is an overhead
of performance. It is advantageous to tolerate the defect and increase accessibility
[3]. One of the problems in cloud computing systems is the vulnerability to fail-
ures. In fact, the availability of the entire system could be compromised whenever a
single node crashes [4]. However, their distribution gives the means of enhancing the
system’s reliability. A tolerant fault system is a configuration that prevents an unin-
tended problem from occurring in a computer system. Fault tolerance means that,
even with failures that cause errors in the results within the system, the provision of
expected services is preserved [5].
Most decisions are made somewhere in the middle, and human decision-makers
can be supported and enhanced by the use of a system of support for decisions
in these cases [6]. In the event of replication or migration of common data blocks
at arbitrary chip sites, a directory or broadcast mechanisms are used to search and
ensure consistency, because the requirements for placement in each block are likely to
differ [7]. In many cases, the usability of a storage system depends on its scalability.
When very many data items are to be stored, or the number of requests for store
exceeds autonomous systems, the logical choice is for the data to be distributed
across multiple physical computers. The logical architectural choice. Replication
should be used [8] if comparatively few data items are used for many requests.
It is desirable to replicate the stock and customers data at these locations since
it provides quick access to local replicas and supports disaster survival in instances
where all physically located machinery crashes [9]. Initially, the Personalized Search
team built a customer side replication mechanism on the big table to ensure that all
replicas were eventually consistent. A replication subsystem that is integrated into
servers is now used in the current system [10]. They have developed data access
pattern replication strategies [11]. Replication of cascading worked well to reduce
the latency of access and rapid replication worked well when the main objective is a
reduction in consumption of bandwidth, but it also wastes a great deal of storage [12].
The study [13] proposed the dynamic model replication strategy. The architecture
they use is peer to peer architecture and in this strategy the replication decision is
taken decentrally. This strategy is limited by the fact that the strategy considers it
to be possible to create replicas without limitation, but in practice it is not possible.
Another overhead is the need to invoke replica location service every time the replica
is created [14]. Of these, we concentrate on data loss and cloud power consumption.
Hardware failures are more likely because of a large number of cloud computing
nodes than non-trivial based on hardware failures statistical analysis in [15]. Some
A Hybrid Multi-user Based Data Replication … 93
failures in the hardware can damage disk node data. Thus, data-intensive applications
that run cannot successfully read data from disks.
2 Related Works
The study [16] proposed a replication technique based on the cloud metrics using
a network bandwidth and the technique was known as the replication based on the
Bandwidth Hierarchy. This strategy for replication uses network-level data access
data. This technique is aimed mainly at reducing time for data access by preventing
network congestion. The site is divided into various areas. One step is to prevent
data duplication and the other to replicate only popular files [17]. Gudeme et al. [18]
offers the branch replication scheme. replication system (BRS). BRS’ architecture
is inherently hierarchical. The data replication system has an important role to play
in the management of grid data [19, 20]. Traditional data replication systems require
very large storage requirements. BRS stresses the optimal consumption of storage.
Contrary to standard replication data for the entire replica on each site, the BRS only
stores the subsection of the replica and it is also possible to have parallel access to
that subsection. This enhances the performance of data access [21–23]. The results of
simulations demonstrate that the BRS system offers better data access performance
and scalability than other systems such as hierarchical replication systems and data
replication schemes directed from server. For read and write operations, the branch
replication scheme is better than the hierarchical replication scheme for all file sizes
[24]. A dynamic replication strategy was proposed to put replicas in hierarchical
data. This scheme is based on the popularity of the file and is named as the PBRP
strategy.
The most commonly accessed file is identified by higher weight value and repli-
cated to appropriate locations for load balance purposes. It demonstrates that LALW’s
work runs in the meantime in terms of efficient network use similarly as LFU
optimizers. Yogendra Naidu et al. [25] proposed an enhanced LALW strategy, the
dynamic data replication concept in data grids, known as enhanced last access largest
weight (ELALW). The study [26] explores cost-performance compromises between
replicated storage systems and erasure-encoded ones. In [27] analyzed the data repli-
cation placement mechanism and developed a heuristic algorithm for the placement
of the data replica. The simulation assesses whether or not the algorithm performs
better in a storage environment. The distributed and replicated, transparent, dynamic
provable data possession (DPDP) for customers has been developed by Alshammari
et al. [28]. It enables the cloud storage provider (CSP), where they can conceal the
internal structure from their customers, manage the resources with flexibility while
still providing t he customer with proven services. And, this work also uses persis-
tent ranked authenticated skip lists to create a dynamic version control system with
optimal complexity that is centralized and distributed.
94 V. Devi Satya Sri and S. Vemuru
3 Proposed Model
Data replication could happen when the same data is saved on several storage devices
or replication, if the same computer work is carried out repeatedly. It is the process
of automatically distributing and maintaining synchronous distribution of copies of
data and database objects among SQL Server instances. Replication is the process of
information sharing, to improve reliability, fault tolerance and accessibility between
the redundant resources, like software or hardware parts. If multiple storage devices
or data replication are used for the same data r eplication, it can be if the same computer
job is carried out many times. Secure information sharing is a difficult problem in
this type of environment. There are two main data of replication protocols: active
replication in which all the replica processes simultaneously concur with all input
message. The owners of different sources have different policies regarding access to
and dissemination of data they hold. The database research community has focused
on passive replication i n which only one replicate processes all input messages and
regularly transfers their current condition to the other replicas for consistency. Data
distribution and replication offer opportunities to improve performance by running
and loading parallel queries and by increasing data availability. Data is often repli-
cated in the distributed database system to increase reliability and accessibility, thus
enhancing their reliability. From Fig. 1, initially, different virtual machines are taken
as input to each user. Since, each user has k number of virtual machines, each user’s
data is partitioned by using the data partitioning algorithm. Each data part in the data
partitioning algorithm is given to integrity model before storing into the VM.
Fig. 1 Proposed Multiple
data partitioning load
balancing parameter
selection
VM_1 VM_2…… VM_n
Amazon AWS
U-1 U-2 U-3
U-1(D) U-2(D) U-n(D)
Data partitioning
Algorithm
Data partitioning
Algorithm
Data partitioning
Algorithm
Cloud data security Data security
framework
VM_1 VM_2…… VM_n
A Hybrid Multi-user Based Data Replication … 95
Blockwise Data Replication Algorithm
Input: clou data files
Output: data files with user access policies.
Procedure:
1. To each file in the cloud user data files
2. Partition the data into k blocks
3. To each block B(i) in k block each with 1024 bits
4. Do
5. Let V_ID be the cloud virtual machine ID with available data zones η.
6. Compute user’s access policy using algorithm 2 as U_P(VM_ID,B(i)).
7. Compute each user’s secret nonce by using the cyclic group parameters as
Let Zr, G1, G2 are randomized cyclic group parameters with generator a.
GauDist(a) = k(1 − k)a,a = 0,1,2,...
UniDist(a) = 1/(r1 − r2) for r1 < a < r2
CyclicElement p = bilinearpair(Zr, GauDist(a));
PrtKey.g =bilinearpair(G1,σ
2
uniDist(a)Un1Dist(a));
PubKey.gp = bilinearpair(G2,a);
MastKey.p = bilinearpair(G2,a);
MastKey.g_alpha = bilinearpair(PubKey.gp,(p)Zr ;
PubKey.h = bilinearpair(PK.g,(P)Zr);
PubKey.g_halpha = bilinearpair(PubKey.g,MastKey.g_alpha);
8. Save each block in the η(VM_ID, B(i), U_P) by using the user’s access
policy.
9. Replicate the block to each VM in the VMList
10. done
In this algorithm, each user’s data file is partitioned and its corresponding block is
replicated to different virtual machines for recovery purpose. In the steps 1–3, each
cloud user’s data file is taken as input and partitioned into blocks with 1024 bits size.
In the steps 4–7, each block in the k blocks is used to compute user’s access policy
and the user’s secret nonce by using the cyclic group metrics. In the step 8, each
block is replicated to multiple virtual machines using zone list. In this step, three
parameters such as virtual machine ID, block data and user’s access policy are used
to replicate the each block in multiple virtual machines. Finally, in the step 9 each
block is replicated in the available virtual machine.
Algorithm 2: User Access Policy Generator
1: Initialize secret key K.
2: Partition the input data M into blocks with size 8..
3: while(len(M)>0).
Do
If(len(M)<8).
Pad message with sequence of …0,000,001;
96 V. Devi Satya Sri and S. Vemuru
else.
Perform block processing using each block partition;
Done
4: BlockProcessing
Divide the block into 32 bit size sub-blocks for non-linear transformation in
the proposed model;
SP[]=BlockPartition[S/32];
For i =0 to len(SP)
Do
While(r<NR) // r current round Do
Do
Perform Subblock Processing(SP[i])
Done
Done
5: Sub block processing.
For each byte in SP[i].
Do
mat_y = |N| · e−|Σ
K−μ|/ρ
2ρ; ρ > 0
η = Norm(mat_y)
gdf(η) = λαxα−1e−λx
⎡(α) ,for x > 0
h1 = sp(i);
h2 = f(sp[i] = log (
λe−λ(sp[i]−τ
(1 + e−λ(sp[i]−τ)2 ∗ gdf(η).mean )
h3 = bytes(mat_y)
H[i]= h1∧h2∧h3
Done
6: H=Concat(H0||H1||||H2||…..||Hn||);
In the Algorithm 2, each user’s access control policy is updated by using the
policy generator. In this algorithm, a secret key and input data is partitioned to find
the block wise access control as shown in steps 1–3. In the step 4, data each block
is sub-partitioned to compute the access policy in the step 5. In the step 5, each
sub-block partition is used to compute the user’s access policy by using the hash
value.
A Hybrid Multi-user Based Data Replication … 97
Fig. 2 Performance analysis
of proposed data replication
model to the conventional
models on cloud storage data
(VM-2)
0
1
2
3
4
5
6
Response me(sec)
Data parons
PDDRA
Dynamicweighted(DWDR)
linearDRA
Proposed
Fig. 3 Performance analysis
of proposed data replication
model storage space to the
conventional models on
cloud storage data (VM-1
and VM-2)
0
5
10
15
20
25
30
35
Storage space
Data parons
PDDRA
Dynamicweighted(DWDR)
linearDRA
Proposed
4 Experimental Results
Experimental results are performed on the data cloud computing environment with
user’s datasets. In this study, Amazon AWS cloud server is used to find the block
wise replication process in the available cloud virtual machines. In this experimental
results, each user’s machine learning patterns are used as input data for replication
process. These patterns are derived from the medical databases using filtered based
classification models (Figs. 2 and 3; Tables 1, 2 and 3).
5 Conclusion
In this work, a hybrid data replication-based multi-user access control mechanism is
designed and implemented on the cloud servers. Most of the conventional models are
difficult to recover the user’s machine learning patterns due to high computational
cost and memory constraints. In this work, a hybrid data partitioning based replication
model is implemented on the user’s machine learning patterns for data recovery and
decision-making process.
98 V. Devi Satya Sri and S. Vemuru
Table 1 Performance analysis of proposed data replication model to the conventional models on
cloud storage data (VM-1)
DataSize PDDRA Dynamic weighted (DWDR) linearDRA Proposed
USER-1
DataSize-1 KB 3.96 4.42 3.33 2.3
DataSize-2 KB 4.65 4.65 3.35 2.35
DataSize-3 KB 4.32 4.54 3.18 2.39
DataSize-4 KB 4.13 3.8 3.28 2.36
DataSize-5 KB 4.73 4.4 3.22 2.33
DataSize-6 KB 4.33 4.32 3.38 2.34
DataSize-7 KB 4.53 3.87 3.15 2.25
DataSize-8 KB 4.2 4.44 3.13 2.31
DataSize-9 KB 3.95 3.85 3.38 2.25
DataSize-10 KB 3.83 4.52 3.32 2.36
Table 2 Performance analysis of proposed data replication model to the conventional models on
cloud storage data (average of all VMs)
DataSize-1 KB 4.23 4.26 3.29 2.39
DataSize-2 KB 4.13 4.6 3.43 2.27
DataSize-3 KB 4.52 4.55 3.18 2.36
DataSize-4 KB 4.61 4.32 3.16 2.36
DataSize-5 KB 4.12 4.73 3.17 2.31
DataSize-6 KB 4.83 3.96 3.28 2.35
DataSize-7 KB 4.34 4.26 3.15 2.34
DataSize-8 KB 4.83 4.42 3.18 2.33
DataSize-9 KB 4.22 3.98 3.15 2.4
DataSize-10 KB 4.09 4.67 3.16 2.26
Table 3 Performance analysis of proposed data replication model storage space to the conventional
models on cloud storage data (VM-1 and VM-2)
DataSize PDDRA Dynamic weighted (DWDR) linearDRA Proposed
VM-1
DataSize-1 KB 23.79 22.8 20.72 17.28
DataSize-2 KB 27.36 29.96 27.06 18.66
DataSize-3 KB 21.82 21.41 26.75 16.3
DataSize-4 KB 27.3 23.85 27.18 16.77
DataSize-5 KB 23.92 28.74 22.1 16.51
DataSize-6 KB 25.6 29.15 20.23 18.45
DataSize-7 KB 24.58 28.49 27.24 16.92
(continued)
A Hybrid Multi-user Based Data Replication … 99
Table 3 (continued)
DataSize PDDRA Dynamic weighted (DWDR) linearDRA Proposed
DataSize-8 KB 23.63 27.26 20.91 16.22
DataSize-9 KB 28.23 26.37 25.17 18.09
DataSize-10 KB 20.05 24.4 22.92 18.58
VM-2
DataSize-1 KB 24.15 29.73 28.49 17.25
DataSize-2 KB 26.66 27.36 20.55 17.22
DataSize-3 KB 26.83 24.72 23.05 16.47
DataSize-4 KB 27.04 20.17 27.05 18.17
DataSize-5 KB 27.07 23.1 29.02 17.66
DataSize-6 KB 24.57 24.58 21.78 16.46
DataSize-7 KB 21.63 22.99 22.87 16.45
DataSize-8 KB 28.86 20.3 21.07 18.61
DataSize-9 KB 21.88 20.44 26.36 16.05
DataSize-10 KB 22.28 22.88 29.9 18
References
1. Casas I, Taheri J, Ranjan R, Wang L, Zomaya AY (2017) A balanced scheduler with data reuse
and replication for scientific workflows in cloud computing systems. Futur Gener Comput Syst
74:168–178. https://doi.org/10.1016/j.future.2015.12.005
2. Alami Milani B, Jafari Navimipour N (2016) A comprehensive review of the data replication
techniques in the cloud environments: major trends and future directions. J Netw Comput Appl
64:229–238. https://doi.org/10.1016/j.jnca.2016.02.005
3. Mokadem R, Hameurlain A (2020) A data replication strategy with tenant performance and
provider economic profit guarantees in Cloud data centers. J Syst Softw 159:110447. https://
doi.org/10.1016/j.jss.2019.110447
4. Gopinath S, Sherly E (2018) A dynamic replica factor calculator for weighted dynamic repli-
cation management in cloud storage systems. Procedia Comput Sci 132:1771–1780. https://
doi.org/10.1016/j.procs.2018.05.152
5. Gill NK, Singh S (2016) A dynamic, cost-aware, optimized data replication strategy for hetero-
geneous cloud data centers. Futur Gener Comput Syst 65:10–32. https://doi.org/10.1016/j.fut
ure.2016.05.016
6. Mansouri N, Mohammad Hasani Zade B, Javidi MM (2020) A multi-objective optimized
replication using fuzzy based self-defense algorithm for cloud computing. J Netw Comput
Appl 171:102811. https://doi.org/10.1016/j.jnca.2020.102811.
7. Mansouri N, Javidi MM (2018) A new Prefetching-aware data replication to decrease access
latency in cloud environment. J Syst Softw 144:197–215. https://doi.org/10.1016/j.jss.2018.
05.027
8. Prajapati P, Shah P (2020) A review on secure data deduplication: cloud storage security issue.
J King Saud Univ Comput Inf Sci. https://doi.org/10.1016/j.jksuci.2020.10.021
9. Helali L, Omri MN (2021) A survey of data center consolidation in cloud computing systems.
Comput Sci Rev 39:100366. https://doi.org/10.1016/j.cosrev.2021.100366
10. Milani BA, Navimipour NJ (2017) A systematic literature review of the data replication tech-
niques in the cloud environments. Big Data Research 10:1–7. https://doi.org/10.1016/j.bdr.
2017.06.003
100 V. Devi Satya Sri and S. Vemuru
11. Briones A, Mallorquí A, Zaballos A, de Pozuelo RM (2021) Adaptive and aggressive transport
protocol to provide QoS in cloud data exchange over Long Fat Networks. Futur Gener Comput
Syst 115:34–44. https://doi.org/10.1016/j.future.2020.08.043
12. Mansouri Y, Prokhorenko V, Babar MA (2020) An automated implementation of hybrid cloud
for performance evaluation of distributed databases. J Netw Comput Appl 167:102740. https://
doi.org/10.1016/j.jnca.2020.102740
13. Setlur AR, Nirmala SJ, Singh HS, Khoriya S (2020) An efficient fault tolerant workflow
scheduling approach using replication heuristics and checkpointing in the cloud. J Parallel
Distrib Comput 136:14–28. https://doi.org/10.1016/j.jpdc.2019.09.004
14. Skakun S, Vermote EF, Artigas AES, Rountree WH, Roger J-C (2021) An experimental sky-
image-derived cloud validation dataset for Sentinel-2 and Landsat 8 satellites over NASA
GSFC. Int J Appl Earth Obs Geoinf 95:102253. https://doi.org/10.1016/j.jag.2020.102253
15. Arora S, Bala A (2021) An intelligent energy efficient storage system for cloud based big data
applications. Simul Model Pract Theory 108:102260. https://doi.org/10.1016/j.simpat.2020.
102260
16. El-Shamy AM, El-Fishawy NA, Attiya G, Mohamed MAA (2021) Anomaly detection and
bottleneck identification of the distributed application in cloud data center using software-
defined networking. Egyptian Informatics Journal. https://doi.org/10.1016/j.eij.2021.01.001
17. Cha J, Singh SK, Kim TW, Park JH (2021) Blockchain-empowered cloud architecture based on
secret sharing for smart city. J Inf Secur Appl 57:102686. https://doi.org/10.1016/j.jisa.2020.
102686
18. Gudeme JR, Pasupuleti SK, Kandukuri R (2021) Certificateless multi-replica public integrity
auditing scheme for dynamic shared data in cloud storage. Comput Secur 103:102176. https://
doi.org/10.1016/j.cose.2020.102176
19. Bello SA et al (2021) Cloud computing in construction industry: use cases, benefits and
challenges. Autom Constr 122:103441. https://doi.org/10.1016/j.autcon.2020.103441
20. Mazzola T (2020) Cloud dimensions from Jack Rabbit II field trial data. Atmos Environ
237:117614. https://doi.org/10.1016/j.atmosenv.2020.117614
21. Nachiappan R, Javadi B, Calheiros RN, Matawie KM (2017) Cloud storage reliability for big
data applications: a state of the art survey. J Netw Comput Appl 97:35–47. https://doi.org/10.
1016/j.jnca.2017.08.011
22. Li C, Song M, Zhang Q, Luo Y (2021) Cluster load based content distribution and speculative
execution for geographically distributed cloud environment. Comput Netw 186:107807. https://
doi.org/10.1016/j.comnet.2021.107807
23. Levitin G, Xing L, Dai Y (2018) Co-residence based data vulnerability vs. security in cloud
computing system with random server assignment. Eur J Oper Res 267(2):676–686. https://
doi.org/10.1016/j.ejor.2017.11.064
24. Torkura KA, Sukmana MIH, Cheng F, Meinel C (2021) Continuous auditing and threat detec-
tion in multi-cloud infrastructure. Comput Secur 102:102124. https://doi.org/10.1016/j.cose.
2020.102124
25. Yogendra Naidu P, Rajalakshmi D, Muthusundari S, Berlin MA, Manikandan K (2021) Cost
efficient tool for cloud storage services. Mater Today: Proc. https://doi.org/10.1016/j.matpr.
2020.10.868.
26. Mohamed MF, El-Gayyar M, Shabayek AE-R, Nassar H (2018) Data reduction in a cloud-
based AMI framework with service-replication. Comput Electr Eng 69:212–223. https://doi.
org/10.1016/j.compeleceng.2018.02.042
27. Fu S, He L, Liao X, Huang C (2016) Developing the Cloud-integrated data replication frame-
work in decentralized online social networks. J Comput Syst Sci 82(1, Part B):113–129. https://
doi.org/10.1016/j.jcss.2015.06.010
28. Alshammari MM, Alwan AA, Nordin A, Abualkishik AZ (2018) Disaster recovery with
minimum replica plan for reliability checking in multi-cloud. Procedia Comput Sci 130:247–
254. https://doi.org/10.1016/j.procs.2018.04.036
Leveraging the Goldfinger Attack
in Blockchain Based on the Topological
Properties
Arcel Kalenga Muteba and Kingsley A. Ogudo
Abstract In this paper, we provide a new approach to modeling and analyzing the
Goldfinger attacks in blockchain networks, based on the topology of a peer-to-peer
network; this paper studied the impact of the Goldfinger well known as 51% attacks
in the case of the ring, mesh and fully connected topology. The outcome of simulation
has proven that in the case of fully connected topology, the attacker node with 501
(hash/s) has found 155 blocks in 60 s and the rest of the nodes, on average, 58 per
node; this gives the control of the balance to the blocks of node J with 131 balance and
43 balance on average for nine nodes, in practice, the node with more than 50% will
monopolize the balance. The Goldfinger occurred very worst in the fully connected
topology and monopolized the network because of topologically connections; the
attacker in fully connected topology is a direct and duplicated connection with the
nine nodes. In terms of connection duplicated bidirectional or directional, the fully
connected topology presents a hard fork compared to ring and mesh topology after
the attack.
Keywords P2P network ·Bitcoin ·Network topologies ·51% attack
1 Introduction
Blockchain-based applications are springing up, covering numerous fields, including
financial services, the Internet of Things (IoT), and so on. Cryptocurrency has sparked
extraordinary interest since it is a new type of currency and a disruptive and inventive
payment method. However, there are still many challenges in blockchain technology,
such as scalability and security problems, waiting to be overcome. On the other hand,
investors bear the risks associated with each transaction, as fraudsters utilize more
A. K. Muteba (B) · K. A. Ogudo
Department of Electrical and Electronics Engineering Technology, University of Johannesburg,
Johannesburg, South Africa
e-mail: arcelkaleng@gmail.com; 219120928@student.uj.ac.za
K. A. Ogudo
e-mail: kingsleyo@uj.ac.za
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_10
101
102 A. K. Muteba and K. A. Ogudo
comprehensive techniques. This paper presents a comprehensive overview of the
Goldfinger attack on blockchain, the implementation and modeling of the attacks
into a peer-to-peer network, and second drafts. It analyzes the impact of the 51%
attacks in the ring, mesh, and full connected topologies based on the following
metrics, blocks found, maximum fork length of the network, and the balance gain.
2 Study Review
In recent years, more research has focused on understanding the cyber criminality
in cryptocurrencies. The following are some propositions to understand, model, and
simulate the cyberattacks in cryptocurrencies. A first model for full-scale simulation
of the bitcoin peer-to-peer network at low r untimes was proposed in [1], as well as
an examination of a network partitioning assault [2]. The validation results show that
key metrics from the simulated and real-world networks are very similar [1]. The
Bitcoin peer-to-peer network was robust to a partitioning assault involving less than
6000 bots and a period of a few hours. However, more resourceful attackers must
be taken into account. A possible assault on bitcoin’s decentralized network design
was conducted, and a data-driven analysis of bitcoin presented probable assaults
based on the network’s spatial and temporal properties. Bitcoin is subject to spatial,
temporal, and logical partitioning assaults, with enhanced attack feasibility due to
network dynamics, and backs up the findings by simulating attack scenarios and the
ramifications for bitcoin [3]. The authors proposed the EREBUS attack in [4], which
splits the bitcoin network without any route modifications, making it undetected
by control-plane and even data-plane detectors. EREBUS transforms the adversary
into a natural man-in-the-middle network of all the peer connections of one or more
targeted bitcoin nodes by patiently influencing the peering decisions of one or more
targeted bitcoin nodes [4].
3 Contribution of the Study
The primary contribution of this study is to implement and model the goldfinger
attacks know as the 51% attacks into a peer-to-peer network. Secondly, to draft and
analyze the impact of the 51% attacks in the ring, mesh, and full connected topologies
based on the following metrics: blocks found, maximum fork length of the network,
and the gain of the balance
Leveraging the Goldfinger Attack in Blockchain Based … 103
4 Methodology and Design
This paper proposes a simulation approach analysis to simulate the goldfinder 51%
attacks in a peer-to-peer network. To build this model blockchain simulator will be
used [5]. This software will be used to construct network topologies, simulate the
behavior of real nodes, and measure the results. Our simulator is based on the bitcoin
prototype. A behavioral analysis of three different network topologies will be drafted,
starting with fully connected, mesh, and ring topologies.
4.1 Peer-To-Peer Network Connectivity
Every node in a peer-to-peer architecture is directly linked to another node. A peer
refers to every computer node. Each peer both delivers and receives services from
other peers. No central server is available [6]. A mesh topology has a point-to-point
link that connects each device to every other device on the network. The link is
exclusively used to carry data between the two connected devices. If the network
has n devices, each device must be connected to (n−1) network devices. In a mesh
topology of n devices, the number of linkages is n (n−1)/2 [7, 8].
Each device in a ring topology is linked to the devices on each side. A gadget
features two dedicated point-to-point links, one on each side. If a device wishes to
communicate data to another device, it does so in one way. Each device in a ring
topology has a repeater, which forwards data to the intended device until it is received
[9]. Finally, a completely linked network is a mesh network with all nodes [10].
5 Implementation
5.1 General Network Before the Goldfinger Attack
We deployed a network composed of 10 nodes connected directional and bidirectional
communicational from one node to nine nodes, using alphabetical denotation from
A to J represented with different colors for the ring and full connected topology.
However, the mesh topology is denoted with random names. The following metrics
parameters are defined for all the topologies (Table 1).
The three network topologies are displayed in Figs. 1, 2, and 3 in the following
section.
104 A. K. Muteba and K. A. Ogudo
Table 1 Parameter of the
network Parameter Value
Power (Hash/s) 100 (Hash/s)
Latency 100 (ms)
Block seize 1,000,000
Difficulty 0.01
Downlink 10–100 (MBps)
Uplink 0.5–10 (MBps)
Fig. 1 Full connected
Fig. 2 Ring topology
Fig. 3 Mesh topology
Leveraging the Goldfinger Attack in Blockchain Based … 105
Fig. 4 Comparison of
topologies before attack
5.2 Results of General Network Before the Goldfinger Attack
The analysis and discussion for this implementation are based upon the following
metrics, the total blocks found, the balance, and the maximum fork length. As high-
lighted in Fig. 4, the results before the attack occurred in the network, in a period
of 60 s, the mesh topology found 682 blocks, followed by fully connected with 646,
and ring 610 blocks found, topologically in peer-to-peer network mesh topology can
broadcast more blocks.
In practice, bitcoin uses mesh topology; it is a random connection. The fully
connected topology presents a total balance of 504 in 60 seconds, then mesh with
408 and ring 379. The maximum fork length in a period of 60 s, this disagreement
within the blocks over speed, block seize, and transaction fee, indicated that the
ring topology presents a soft maximum length with 811 probably the double of full
connected.
5.3 Implementation of the Goldfinger Attack
To simulate the Goldfinger in the network depend directly on the hash rate and the
power of the nodes, we powered the last node with 501 (Hash/s) is, representing more
than 50% of the power in the network, this is for node J in Ring and full connected
topologies and Austin in a mesh topology, and the rest of the nodes 100 (Hash/s),
and the minimum downlink set to 10 MBps and maximum to 100 MBps. The uplink
was set respectively to 0.5 MBps for the minimum and 10 MBps for the maximum.
The difficulty of adding a new block to the chain is 0.01, and the latency is set to
100 ms, as shown in the attached Table 2.
106 A. K. Muteba and K. A. Ogudo
Table 2 Parameter of the
attack Parameter Value
Power (Hash/s)
Attacker power (Hash/s)
Latency
Block seize
Difficulty
Downlink
Uplink
100 (Hash/s)
501 (Hash/s)
100 (ms)
1,000,000
0.01
10–100 (MBps)
0.5–10 (MBps)
6 Results
The results are shown in Fig. 5 for the fully connected topology, Fig. 6 for ring
topology, and mesh topology is on Fig. 7, and the comparison of the three topologies
is in Fig. 8.
Fig. 5 Attack on full
connected
Fig. 6 Attack on ring
Leveraging the Goldfinger Attack in Blockchain Based … 107
Fig. 7 Attack on mesh
Fig. 8 Comparison after
attack
7 Discussion and Analysis
In the case of full connected topology, the attacker node with 501 (hash/s) has found
155 blocks in 60 s and the rest of the nodes, on average, 58 per node, which gives
them control of the balance to the blocks of node J with 131 balance and 43 balance
on average for nine nodes as indicated in Figs. 5 and 8. In-ring and mesh topology,
we got respectively for the attackers 115 and 121 balance and, on average, 35 and
39 for the rest of the nodes, as shown in Figs. 6 and 8.
The goldfinder occurred very worst in the full connected topology and monop-
olized the network because topologically, the attacker in full connected topology
is a direct and duplicated connection with the nine nodes. In terms of connection
duplicated bidirectional or directional, the full connected topology presents a hard
fork compared to ring and mesh topology after the attack. The maximum fork length
in a period of 60 s, this disagreement within the blocks over speed, block seize, and
transaction fee occurred 43 times for full connected, 56 times in ring, and 53 for
mesh topology.
108 A. K. Muteba and K. A. Ogudo
8 Conclusion
In this work, we proposed the implementation and model of the goldfinder known as
the 51% attacks into a peer-to-peer network. Secondly, we drafted and analyzed the
impact of the 51% attacks in the ring, mesh and full connected topology based on
the following metrics, blocks found, maximum fork length of the network, and the
gain of the balance. The model was successfully implemented using a blockchain
simulator; the results of the simulation show that in the case of a fully connected
topology, the attacker node with 501 (hash/s) found 155 blocks in 60 s, while the rest
of the nodes found 58 blocks on average per node, giving control of the balance to the
blocks of node J with 131 balance and 43 balance on average for 9 nodes. In practice,
the node possessing more than 50% of the balance will dominate it. The attacker in
the full connected topology is a direct and duplicated connection with the 9 nodes.
The goldfinder occurred badly in t he fully connected topology and monopolized the
network because of the topological connection. For further work, understanding all
the metrics of the peer-to-peer network will be considered for implementing the same
attack and its impacts.
References
1. Kim TW, Zetlin-Jones A (2019) The ethics of contentious hard forks in blockchain networks
with fixed features. Front Blockchain:9
2. Banerjee S, Das D, Biswas M, Biswas U (2020) Study and survey on blockchain privacy
and security issues. In: Cross-industry use of blockchain technology and opportunities for the
future. IGI Global, pp 80–102
3. Neudecker T, Andelfinger P, Hartenstein H (2015) A simulation model for analysis of attacks
on the Bitcoin peer-to-peer network. In: The 2015 IFIP/IEEE international symposium on
integrated network management (IM). IEEE, pp 1327–1332
4. Fan W, Chang SY, Zhou X, Xu S (2021) Conman: a connection manipulation-based attack
against bitcoin networking. In: 2021 IEEE conference on communications and network security
(CNS). IEEE, pp 101–109
5. https://simewu.com/blockchainsimulator/index.html
6. Fox G (2001) Peer-to-peer networks. Comput Sci Eng 3(3):75–77
7. Bisht N, Singh S (2015) Analytical study of different network topologies. Int Res J Eng Technol
(IRJET) 2(01):88–90
8. Thai CDT, Lee J, Prakash J, Quek TQ (2018) Secret group-key generation at the physical layer
for multi-antenna mesh topology. IEEE Trans Inf Forensics Secur 14(1):18–33
9. Puelles L, Alonso A, García-Calero E, Martínez-de-la-Torre M (2019) Concentric ring
topology of mammalian cortical sectors and relevance for patterning studies. J Comp Neurol
527(10):1731–1752
10. Tsujimoto T, Shindo T, Kimura T, Jin’no K (2012) A relationship between network topology
and search performance of PSO. In: 2012 IEEE congress on evolutionary computation. IEEE,
pp 1–6
Bitcoin Transaction Computational
Efficiency and Network Node Power
Consumption Prediction Using
an Artificial Neural Network
Arcel Kalenga Muteba, Kingsley A. Ogudo, and Espoir M. M. Bondo
Abstract This paper develops and discusses the predictability of power consump-
tion of the Bitcoin networks using Artificial Neural Networks (ANN) machine
learning algorithm and solving the computational problem of the Bitcoin mining
process. It discussed its impacts on energy consumption in the crypto mining process.
In this paper, we used data sets for Bitcoin historical information for the training and
testing of the ANN algorithm. With the help of Python libraries, the data filtration
process was done. Python has provided the best feature for data analysis and visual-
ization. After understanding the data, we trim the data and use the characteristics or
attributes best suited for the model. The implementation of the model is done, and the
result is recorded and analyzed. The results as obtained demonstrated that the use of
Artificial Neural Networks (ANN) machine learning algorithm could approximately
predict the actual electricity consumption of Bitcoin with high accuracy.
Keywords Bitcoin ·Artificial Neural Network ·Power consumption
1 Introduction
Among the leading technologies of modern times, blockchain uses a large amount
of electricity [1, 2]. For decades, Bitcoin has been used in central processing units
(CPUs) such as laptops and desktop computers. Early miners can get Bitcoin with
low-cost hardware and a personal computer. The growing interest in Bitcoin mining
A. K. Muteba (B) · K. A. Ogudo
Department of Electrical and Electronics Engineering Technology, University of Johannesburg,
Johannesburg, South Africa
e-mail: 219120928@student.uj.ac.za; arcelkaleng@gmail.com
K. A. Ogudo
e-mail: kingsleyo@uj.ac.za
E. M. M. Bondo
Engineering Research and Development BOND’AF, Paris, France
e-mail: espoirbondo@bondaf.com
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_11
109
110 A. K. Muteba et al.
Fig. 1 Bitcoin energy
consumption
has led miners to discover that graphics cards can implement algorithms more effi-
ciently and help add cryptocurrency. Graphic cards were later replaced by system-
atic port gate systems called Field Programmable Gate Array (FPGAs). Users may
adjust their circuits after being manufactured and then integrated system-specific
circuits (ASICs). Bitcoin mining ASICs have changed from place to place; as mining
equipment has become more sophisticated, the mining industry has agreed to join
mining pools to s pread profits and speed up transactions; this is very expensive for
many miners. The ASICs used for Bitcoin mines typically reside in temperature-
controlled data centers with easy access to electricity. The Cambridge Bitcoin Elec-
tricity Consumption Index (CBECI) calculates a hypothetical range that includes a
lower bound (floor) and upper bound (ceiling) estimate. A best-guess estimate is
calculated within the parameters of this range to produce a more realistic value that
approximates Bitcoin’s actual electricity consumption [3] as illustrated in Fig. 1 the
annual Bitcoin energy consumption.
This paper presents the feasibility of predicting the energy use by Bitcoin network
using the Artificial Neural Networks (ANN) machine. We will investigate how histor-
ical data from the Bitcoin network can be used to solve the computer problem
of Bitcoin mining and the impact of various types of synchronization techniques
depending on the power consumption of the crypto mining process.
2 Study Review
Evaluating the environmental impact of Bitcoin and other cryptocurrencies has
become the curiosity of many researchers. Energy-derived emissions from mining
might drive global warming past two °C, according to Mora, C., and Dittmar, L., and
this currency’s annual use is expanding at an exponential rate, reaching a staggering
55 TWh [4]. This is, without a doubt, a severe issue; in the first half of 2018, between
3 and 13 million metric tons of CO2 were released into the atmosphere due to Bitcoin
mining [5]. We learned unsettling information in 2016; the annual energy usage of
Bitcoin mining is estimated to be 3.38 Terawatt Hours in this article (TWh) [7]. This
enormous amount of energy equals Jamaica’s whole annual energy consumption
Bitcoin Transaction Computational Efficiency and Network Node … 111
in 2014. The Bitcoin network’s total energy consumption is more significant than
Ireland’s [6]. The report estimates that by the end of 2018, Bitcoin will consume
0.5% of the world’s electricity. We know that the electricity requirement is due to
intricate computing, and as time passes, more complex problems must be solved.
[6, 7].
Despite these technological advances, Bitcoin mining has not yet been trans-
formed into a highly integrated industry as various Bitcoin mining pools continue
to compete using the Proof of Work (PoW) method [8–10]. If the use of Bitcoin
continues to rise, future generations may be forced to deal with profound implica-
tions. The total amount of money in circulation worldwide is expected to be 11,000
billion U.S. dollars. As a result, the associated energy usage will exceed 4000 GW
[11]. This enormous energy is eight times France’s and doubles the U.S.’s combined
electrical consumption [11]. As a result, Bitcoin may become a climatic burden.
3 Contribution of the Study
This paper firstly gives an understanding regarding the accurate calculation of energy
consumption of cryptocurrency in the network. Secondly, we developed the feasi-
bility of predicting the energy consumption of Bitcoin using the Artificial Neural
Networks (ANN) machine learning. We investigate how historical data from the
Bitcoin network can be used to solve the computer problem of Bitcoin mining and
the impact of various types of synchronization techniques depending on the power
consumption of the crypto mining process.
3.1 Methodology
Our primary methodology is to predict the energy consumption of Bitcoin using the
ANNs; but before we started with the calculation of energy demand of Bitcoin and
then, data collection transactions in the Bitcoin network are stored in the blockchain.
We ran the Bitcoin client in our local machine to get the latest blockchain. After
collecting the blockchain, we parsed it into blocks and transactions. Each block has
its hash value after all data collection is done. We divide the data into three parts:
test data, training data, and cross-validation for machine learning statistics data. The
prediction uses a training data set, and 80% of data is allocated for training and 20%
for testing. Day transactions from 1 August 2019 to 1 June 2021 are included in
the data collection. Although it is more difficult to forecast when there is increased
volatility in the Bitcoin price, ANN machine learning algorithms attempt to predict
with some degree of accuracy (Fig. 2) (Table 1).
112 A. K. Muteba et al.
Fig. 2 Data flow
Table 1 Calculation
parameters Parameters
Mining difficulty daily
Miner fees, daily
Hash rate, daily
Bitcoin market price
Electricity cost
Power usage effectiveness
Average block time
Parameters to consider during the calculation of energy consumption
Description of Bitcoin mining
Mining is the process that Bitcoin uses to create new Bitcoins by solving complex
math problems that verify transactions in the currency. The miner receives the reward
amount of Bitcoin when a Bitcoin is successfully mined. Parameters that determine
the power consumption in cryptocurrency are hardware computing power, network
hash rate, thermal regulation for the hardware, and the mining difficulty. We can as
well determine the maximum obligation requirement of the PoW blockchain mining
process, assuming that honest and sensible miners who benefit from mining only to
Bitcoin Transaction Computational Efficiency and Network Node … 113
participate in the profit margin of the mining process are profitable as long as the
expected revenue from the mines exceeds the corresponding mining costs [1]. The
total power consumption is provided by Eq. 1.
PT = CP + Br + Tf
Bav + E p
(1)
where PT total power consumption, Br block reward, Tf transaction fees, CP coin
price, Bav avg. block time, and E p min. electricity price. However, the annual energy
consumption triggered by the storage of a fully replicated blockchain can be estimated
with Eq. 2.
ESt
KWh
Year = #Repl ∗ BCSt[GB] ∗ EISt
KWh
year ∗ GB (2)
where ESt is the yearly energy for storing the blockchain, BCSt is the size of the stored
blockchain, EISt is the average energy intensity of holding a unit of data (1 G.B) for
one year, and Repl is the average number of replicas. With an assumption of the
global average electricity price is 0.05 USD/kWh and remains constant [3].
Machine learning
Machine learning enables systems to make decisions independently, without human
intervention. When the machine can scan the data and understand the basic patterns,
these decisions are made. The effect can then be segmented or anticipated using
pattern matching and subsequent research. Louridas and Ebert [12] The field of
machine learning is separated into three categories: supervised, unsupervised, and
reinforced learning; because the proposed method is intended to estimate power
consumption, supervised reading is a better fit because its primary function is to
display the value of the target variable based on predictable variability [13, 14].
4 Results and Discussion
We deployed the model that gives the estimation and predicted the Bitcoin electricity
monthly and cumulative consumption, from January 2019 to May 2021, as shown in
Table 2.
Moreover, the results of the predicted annual power consumption from the same
period are presented in Table 3 and displayed in Fig. 3, and we found the prediction
accuracy of A = 0.978 of the annual average power consumption of Bitcoin.
For the second case, we find that the prediction accuracy of A = 0.945 of the
annual low power consumption of Bitcoin is shown in Fig. 4, and the last point, we
find that the prediction accuracy of A = 0.982 of the yearly high or Pick power as
highlighted consumption of Bitcoin is shown in Fig. 5.
114 A. K. Muteba et al.
Table 2 Cumulative and monthly consumption/TWh
Period Monthly consumption/TWh Cumulative consumption/TWh
January 2019 3.787 80.879
Feb 2019 2.905 82.891
March 2021 3.658 84.452
– – –
April 2021 9.127 235.312
May 2021 9.766 243.89
Table 3 Annual average power consumption
Period Annual power consumption of Bitcoin/TWh Accuracy of ANN
Average 129.21829335 0.978
low 42.214 0.945
High 482.8749232 0.982
Fig. 3 The annual average power consumption of Bitcoin
Fig. 4 The annual low
power consumption of
Bitcoin
The prediction accuracy is dropping down with the increase of feature prediction
parameters. The first direct correlation exists between the mining difficulties with
the energy consumption. As Ethereum switched from Proof of Work to Proof of
Stake, there is a pressing need for Bitcoin to reduce mining difficulty and adapt to a
Bitcoin Transaction Computational Efficiency and Network Node … 115
Fig. 5 The annual at pick
point power consumption of
Bitcoin
different consensus model. Despite its complexity and training time limits, the ANN
model of machine learning provides excellent accuracy, as evidenced by all graphs.
Future work might be to discover other models that can easily predict, for instance,
Deep Neural Networks and Convolutional Neural Networks, to see if changes affect
model accuracy.
5 Conclusion and Future Work
This paper focuses on the predictability approach of Bitcoin energy consumption
using Artificial Neural Networks (ANN). We discovered that using the Artificial
Neural Networks (ANN) machine learning technique, we can estimate the actual elec-
tricity usage of Bitcoin with a high degree of accuracy, but determining the exact value
for the energy consumption of a multitude of open distributed networks is a complex
task because the precise number of participants, the properties of their hardware, and
the effort which they put into mining are unknown. This paper only compares Artifi-
cial Neural Networks (ANN) with the approximated values of CBECI. In the future,
further machine learning models will be compared to confirm the result. Another
approach that could be experimented with is to predict the energy consumption of
various cryptocurrencies.
References
1. Johannes S, Hans U, Gilbert, Robert K (2020) The energy consumption of blockchain
technology: beyond myth
2. Peck ME (2013) The bitcoin arms race is on. Spectrum, IEEE 50(6)
3. Cambridge Bitcoin Electricity Consumption Index (CBECI). https://ccaf.io/cbeci/index
4. Mora C, Rollins RL, Talalay K, Kantar MB, Chock MK, Shimada M, Franklin EC (2018)
Bitcoin emissions alone could push global warming above 2 °C. Nat Clim Change
5. Dittmar L, Praktiknjo A (2019) Could bitcoin emissions push global warming above 2 °C. Nat
Clim Change
6. Hern A (2017) Bitcoin mining consumes more electricity a year than Ireland
7. Huckle S (2018) Bitcoins energy consumption is a concern, but it may be a price worth paying
8. Allied Control—Analysis of large-scale bitcoin mining operations (White Paper)
9. Brito J, Bitcoin (2013) A primer for policymakers
116 A. K. Muteba et al.
10. Shalev-Shwartz S, Ben-David S (2014) Understanding machine learning from theory to
algorithms
11. Fabrico Flipo; The bitcoin and blockchain: energy hogs
12. Louridas P, Ebert C (2016) Machine learning, Paris. The computer society, IEEE
13. Vanita Tonge Buradkar MM (2020) Introduction to machine learning and its applications: a
survey. J Artif Intell Mach Learn Soft Comput
14. Ongsulee P (2017) Artificial intelligence, machine learning, and deep learning. In: Fifteenth
international conference on ICT and knowledge engineering, Thailand
Remote Breast Cancer Patient
Monitoring System: An Extensive Review
Sangeeta Parshionikar and Debnath Bhattacharyya
Abstract The healthcare domain is one of the fastest-growing fields for the Internet
of Things and Artificial Intelligence. The advancement of medical resources is insuf-
ficient to meet the needs of remote patient monitoring and treatment. This issue is
growing increasingly prevalent in developing countries. The convergence of IoT and
AI solves this problem significantly. A remote monitoring system for breast cancer
patients is urgently needed in order to provide effective care to them. This study exam-
ines related research on existing and future technologies for breast cancer detection,
and how the confluence of IoT and AI is leading to the emergence of smart healthcare.
Various breast cancer screening approaches have been briefly addressed, as well as
popular public databases. Following that, issues in remote monitoring system have
been discussed. We also present a case study on remote monitoring system for breast
cancer patients to provide enhanced solution for women in rural areas.
Keywords Healthcare ·Breast cancer ·Remote patient monitoring ·Hospital
management ·IoT ·Wearables ·Artificial Intelligence
1 Introduction
Healthcare sector is moving toward the automation. From surgery to hospital manage-
ment, manual diagnosis to automated diagnosis using CAD tools, Artificial Intelli-
gence and the Internet of Things (IoT) technologies are playing a key role in the
growth of smart healthcare. By incorporating these technologies, the performance
of various domains of medical application such as diagnosis, prognosis, monitoring,
spread control, and assistive systems is improving and still research is going on [1,
2]. Early diagnosis and detection of life-threatening diseases such as cancer, heart
disease, or diabetics is being possible with IoT and subset of AI, deep learning (DL).
S. Parshionikar (B) · D. Bhattacharyya
Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation,
KLEF, Guntur, Andhra Pradesh, India
e-mail: sangeeta05@gmail.com
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_12
117
118 S. Parshionikar and D. Bhattacharyya
Fig. 1 The components of
smart healthcare
Smart
H
Healthcare
Smart
Hospital
Remote
Patient
Monitor-
On
body
Sensors
Emergency
Response
Mobile
Health
Doctor/
Nurse/
Technician
Telemedicine
Though there are conventional cancer therapies like surgery, chemo, and radi-
ology which will remain for next many years. AI is bringing a novel and powerful
tool to fight against cancer [3]. AI will make cancer treatment revolutionized. Still,
there are few limitations at different levels in the healthcare sector while handling AI-
based tools. Those include patient’s data confidentiality, dataset size, classification on
many types of cancers, unregulated training set, etc. Smart healthcare is becoming an
emerging research subject as the globe evolves toward remote monitoring, concur-
rent, real-time detection of most of the diseases. Smart healthcare is divided into
several categories such as telehealth, mobile health, and RPM, all of which refer to
the use of IoT to monitor patients remotely located [4]. The components involved in
smart healthcare can be seen in Fig. 1.
2 Related Work on Breast Cancer
By 2030, breast cancer will be the major cause of mortality among women, outnum-
bering all other diseases. Weight gain, a lack of exercise, hormone changes and medi-
cation, the excess of oral contraceptive pill, stress, and most likely late work sched-
ules are few of the factors that raise the risk of breast cancer [5]. Breast cancer has
a low survival rate since it is often detected late. Early identification and treatment,
according to the World Cancer Report 2020, are the most effective interventions for
breast cancer control [6]. Modern techniques and recent diagnostic advancements
are beneficial in addressing these issues since they are non-invasive and painless and
detect tumor in the minimum time, and as a result, breast cancer can be treated at an
earlier stage than previous approaches.
Remote Breast Cancer Patient Monitoring System … 119
Fig. 2 Screening methods
of breast cancer
2.1 Existing Approaches to Detect Breast Cancer:
Early detection and treatment are the most important factors in a woman’s breast
cancer survival. Breast Ultrasound (BU), mammography, Magnetic Resonance
Imaging (MRI), Computed Tomography (CT), thermography, and the biopsy and
Microwave Breast Imaging are the breast cancer screening procedures that have been
developed. Many existing screenings and the developing technologies are utilized
to diagnose breast cancer in its early stages [7]. The current breast cancer screening
techniques are classified as shown in Fig. 2.
2.2 Popular Datasets Available for Research
This section delves into a detailed examination of public datasets available for the
classification of various breast cancer experiments. Breast Cancer Digital Reposi-
tory (BCDR—Film Mammography (FM), Full Field Digital Mammography (DM)),
Curated Breast Imaging Subset of Digital Database f or Screening Mammography
(CBIS-DDSM), Digital Database for Screening Mammography (DDSM), INBreast,
Breast Cancer Wisconsin (Original) Data Set, Bio-Imaging Challenge 2015 Breast
120 S. Parshionikar and D. Bhattacharyya
Histology (BICBH), Breast Cancer Histopathological Image (BCHI), and Mammo-
graphic Image Analysis Society (MIAS)/mini-MIAS are the eight public datasets
available for classification of breast cancer, as shown in Table 1 (BreakHis) [8].
Exclusive datasets have fewer annotated photos than public datasets. As a result,
the model that was tested on public datasets outperforms the model that was tested
on private datasets. Regardless of database type, grayscale images (mammograms,
ultrasound, and MRI) or colored images (histopathology) are used for breast cancer
classification at the abstract level. Furthermore, the majority of studies concentrated
on binary classification, with only a few focusing on multi-class problems for breast
cancer classification.
2.3 Deep Learning Techniques for Early Diagnosis of Breast
Cancer
Saad Awadh Alanazi et al. suggested a method to detect breast cancer in whole-
slide images automatically. Convolutional neural network (CNN) method is used
to improve the detection by evaluating hostile ductal carcinoma tissue zone. In
this study, three distinct CNN designs are discussed, and their performances are
compared. Author had calculated four performance metrics, namely accuracy, preci-
sion, recall, and F1 score. The employed CNN Model 3 achieved an accuracy of 87
percent, potentially reducing human errors in the diagnosing process.
Accuracy =Number of true predictions(TP + TN)
Total number of predictions(TP + TN + FP + FN) (1)
Precision =Number of true positive(TP)
Sum of number of true postive and false positive(TP + FP) (2)
Recall =Number of true positive(TP)
Sum of number of true positive and false negative(TP + FN) (3)
F1 Score = 2 × precision × recall
precision + recall (4)
Model 3’s five-layer CNN is deeper than Models 1 and 2 and hence proved to be
best among all. A large collection of roughly 275,000, 50 × 50 pixel, RGB picture
patches guided all structures. The use of a secondary database like Kaggle is a major
drawback of this study. For more accurate breast cancer detection outcomes, future
studies should be carried out on primary dataset [8].
Anji Reddy Vaka et al. proposed a new approach for identifying breast cancer
called Deep Neural Network with Support Value (DNNS). Their proposed solution
Remote Breast Cancer Patient Monitoring System … 121
Table 1 The list of publicly available dataset
S. No Dataset name Number of
patients
Number of
Images
Image format Class distribution
1Breast Cancer
Digital Repository
((BCDR)—Film
Mammography
(FM), Full Field
Digital
Mammography
(DM))
1734
1010
724
7315
3703
3612
TIFF
TIFF
TIFF
Benign and
malignant
2Curated Breast
Imaging Subset of
DDSM
(CBIS-DDSM)
6775 10,239 DICOM Normal, benign,
and malignant
3Digital Database
for Screening
Mammography
(DDSM)
2620 10,480 JPEG Normal, benign,
and malignant
4 INBreast 115 410 DICOM
5Breast Cancer
Wisconsin
(Original) Data Set
699 –10 Attributes 458 benign and
241 malignant
6 Bioimaging
Challenge 2015
Breast Histology
(BICBH)
285 285 TIFF Normal, benign,
in situ, and
invasive
7Breast Cancer
Histopathological
Database
(BreakHis)
40×
100×
200×
400×
82 9109
1995
2081
2013
1820
PNG 2480 benign and
5429 malignant
8Mammographic
Image Analysis
Society (MIAS)
–322 (50
micron
resolution)
PGM Normal, benign,
and malignant
9Mini-MIAS –322 (200
micron
resolution)
PGM Normal, benign,
and malignant
is based on a deep neural network’s support value. To improve the performance, effi-
ciency, and quality of photographs, a normalization method is used. Firstly, prepro-
cessing is done on images to remove noise using Gaussian filtering technique. Further,
Histo-Sigmoid Fuzzy Clustering is used to segment brain tumor from the extracted
images. Following histogram function and sigmoid functions are used.
122 S. Parshionikar and D. Bhattacharyya
Hg =
x
Σ
i=1
Hgi(5)
∂ =1
1 + ex (6)
Author has used following pseudo-code for their proposed algorithm. Experiments
have shown that the suggested DNNS is far superior to existing approaches. The
proposed method is proven to be favorable in terms of performance, efficiency, and
image quality, which is critical in today’s medical systems [9].
Support value - based normalized image (SN) = port value ×X − Xmin
Xmax − Xmin
(7)
In order to reduce the breast cancer-related mortality rate, Hassanien et al. looked
at a variety of recently developed models for breast cancer diagnosis and catego-
rization. They have used most common and a big database the CBIS-DDSM. In the
paper, it is inferred that YOLO and RetinaNet are novel models that have recently been
employed and are believed to be more simple than conventional CNN networks in
terms of mass detection and malignancy classification. They achieve better outcomes
and more accurate performance [10].
Salvi proposed [11] a technique that could enable a patient evaluates whether she
is at risk for breast cancer at an early stage, allowing the breast cancer cells to be
removed with adequate therapy. A paper also includes detailed information about IoT
technology combined with the machine learning to assist patients living in distant
areas with limited access to doctors. According to their findings, when a patient
wants to know the condition of a cancer cell, he or she can utilize a thermal imaging
sensor to take cancer cell pictures. The image obtained is delivered as the input to
the deep learning trained model using the Raspberry Pi microcontroller board. The
model uses CNN to analyze the input image whether the image provided by the
patient is normal, benign, or malignant.
Abdelhafiz et al. [12] examined the performances, strengths, and limits of the
most recent CNN applications in evaluating mammography (MG) pictures in great
detail. This study examines contemporary CNN approaches in MG pictures, demon-
strating how developments in DL algorithms produce promising findings that can
assist radiologists and improve diagnosis time. CNN algorithms could be used to
process millions of regular imaging exams, showing probable cancers to radiologists
who undertake follow-up operations. Other techniques such as transfer learning, data
augmentation, batch normalization, and dropout which are all tempting options for
reducing overfitting and increasing the generalization of the CNN model are also
explained in this paper.
The goal of Mashekova et al. study [13] was all about the detection of breast cancer
abnormalities using Infrared Breast Thermography. Paper provided a complete
overview of IBT method. The author of this paper suggested that thermography
is one of the safest and least intrusive breast cancer screening modalities. They also
Remote Breast Cancer Patient Monitoring System … 123
looked at whether this non-contact, low-cost technique has a lot of potential for early
breast cancer detection through a large-scale screening with ongoing monitoring of
questionable individuals. Thermography is used to diagnose breast cancer by identi-
fying certain aspects of breast heat trends over time. Through research, they found the
spectral radiance and wavelength relationship, which is given by Planck’s radiation
law.
F = 2hc2
λ5(e hc
λKT − 1)−1 (8)
where F—radiation power, h—Plank’s constant, c—speed of light in vacuum, and
K—Boltzmann constant.
However, they also emphasized that, at this time, thermography can only be used
as a supplement to mammography because it is so sensitive to the procedure’s condi-
tions and the patient’s overall health. They highlighted an important point: Because
of the unavailability of standard cancer screening methods in rural and distant places,
alternative technologies such as thermal imaging, according to the author (thermog-
raphy), have been developed. The authors built a system that uses the extremely
powerful object identification Faster Region-based CNN, a deep learning approach
to automatically detect and classify breast cancer lesions in mammograms. A total
of 330 mammography images are used in their proposed CAD system, with 121
annotated images being used to train the Faster Region-based CNN network. For the
testing set, the suggested system had a mAP (mean Average Precision) of 0.857 and
performed well in detecting mammographic lesions.
mean Average Precision (mAP) = 1
n
k
Σ
i=1
APi(9)
where n-number of classes and AP—the average precision of class i.
Traditional Faster Region-based CNN has the drawback of failing to recognize
multiple smaller objects. Future work should include the CBIS-DDSM database,
according to the paper.
3 Challenges and Issues in Remote Monitoring System
Internet of Things is steadily revolutionizing the healthcare industry. This is possible
with the use of smart sensors which collect information, actuators, various services
provided by cloud like storing, processing, and analyzing, and short-long range
communication protocols. The Internet of Things has the ability to enable remote
monitoring. This functionality might be beneficial in providing remote monitoring
for patients with breast cancer [14]. The actual implementation in a hospital is still
a long way off. The remote monitoring system has some of the open concerns and
124 S. Parshionikar and D. Bhattacharyya
challenges that need to be addressed more in the future. In this section, few of the
challenges have been mentioned.
1. Scalability–Scalability is a characteristic of healthcare device. It is about how
device responds to the environmental conditions. A device with high scalability is
recommended. It is critical to build a high scalability device. IoT in the healthcare
system comprises various health parameters measuring sensors, medical devices,
actuators, and cloud storage, and all of them communicate and share data using
Internet. There should be homogeneity across healthcare IoT system devices in
order to achieve high scalability [15]. A system with high scalability functions
smoothly and efficiently, utilizing all available r esources. Hence, scalability of
the healthcare IoT system must be managed well.
2. Power Consumption—The main source of power to the majority of IoT devices
is a battery. These battery-based devices have limitation. Once these devices are
installed in the system, it is difficult to change a device’s battery. Hence, high-
capacity battery powered devices are required in the healthcare IoT system. It
is vital to create IoT devices for healthcare that can produce their own power
[11]. Renewable energy can be one of the possible solutions. The healthcare
IoT system integrated with renewable energy will definitely improve the life of
devices. These systems may be able to help alleviate the global energy issue to
some extent. Also, when no sensor readings need to be reported, these devices
save energy by turning on the power-saving mode [15]. Furthermore, if there is
nothing critical to process, they run at a low CPU speed.
3. Data Privacy and Security—Data privacy and security are the major issues in
the IoT systems. Because of millions of devices connected to the Internet, there
is always a threat that sensitive data can be mishandled by attackers or can be
misused. It is a challenging task to provide secured exchange of data between
heterogeneous devices. In order to provide secured interaction, IoT systems must
be equipped with a robust authentication approach [12, 16].
4. Self-Configuration—Self-configuration is a characteristic feature of every IoT
system. This feature allows IoT system to automatically configure, like auto-
matically upgrade software versions or operating systems. It should also give
user a power to integrate features like manual configuration. Manual configura-
tion will allow users to adjust system parameters as per application demand and
circumstances [17].
5. Servicing and Maintenance—Involvement of heterogeneous medical devices and
sensors makes IoT system’s maintenance expensive. As a result, IoT system must
include devices and sensors with low maintenance cost. It is really challenging to
choose such devices and come up with the system which has low maintenance,
repair, and upgrade costs [17, 18].
Remote Breast Cancer Patient Monitoring System … 125
4 Case Study: A Remote Breast Cancer Patient Monitoring
System
Breast cancer has a low survival rate since it is often detected late. Most of breast
cancer patients consult a doctor for the first time when they are in stage 3, and almost
15% of patients visit a doctor when they are in stage 4. Breast cancer is a curable
disease with a better chance of survival if caught early. The most common reason
why women do not get treatment on time is breast tumor causes no pain [11, 19].
Ideally women should get a clinical breast exam done by a doctor every three years.
Following that, an ultrasound scan should be included in a regular check-up. The
facilities like mammography, ultrasound, and check-up by expert are rarely available
in rural areas. A remote breast cancer patient monitoring system will prove to be boom
for remotely located women, and they can easily avail these facilities.
A remote breast cancer patient monitoring system can consist of three main enti-
ties: rural/remote area, local primary health center, and medical expert. Primary health
center (PHC) will be equipped with IoT board like Raspberry Pi with thermal IR
camera and IoT communication technologies. Women from rural areas will approach
local PHC. In initial physical examination, if any symptoms like change in shape
or size of breast, skin dimpling, swelling, or redness are observed, images from IR
camera will be taken. PHC will send this data to the cloud through IoT board for
further processing.
Based on uploaded data on cloud, trained deep learning model will detect whether
the patient is normal or have any malignancy. A notification will go to medical expert,
oncologist, or hospital if malignancy is found. Medical expert will go through all
the details and recommend line of treatment. Along with the treatment procedure,
there would be provision to inform the patients about nearest cancer care center.
The detailed overview of proposed remote breast cancer patient monitoring system
would be the part of our future research paper.
5 Result and Analysis of Existing Techniques
Early identification of breast cancer can be achieved using deep learning algorithms.
There have been researches on the subject that have used different public and exclu-
sive datasets. In recent studies, convolutional neural networks (CNN), faster R-CNN,
and DNNS have been used as DL techniques for breast cancer diagnosis. Table 2
contains a detailed analysis of these studies, including their objective, technique,
dataset type, performance measures, and future scope.
126 S. Parshionikar and D. Bhattacharyya
Table 2 Analysis of deep learning techniques to detect breast cancer
Research/Author Objective Technique used Dataset Performance
parameter
Future scope
Alanazi et al. [8] In WSI
detection of
BC by
examining
hostile ductal
carcinoma
tissue
Convolutional
neural network
(CNN)
Kaggle 162
H&E
Acc = 87% More
accuracy can
be achieved
by carrying
study on
primary data
for breast
cancer
identification
Vaka et al. [9]Detection of
BC using
machine
learning
techniques
New
technique—Deep
neural network
with support
value
8009
histopathology
image samples
from M. G.
Cancer
Hospital and
Research
Acc =
97.21%
P = 97.9%
R = 97.0%
–
Salvi and
Kadam [11]
Detection of
breast cancer
at an early
stage using
IoT
CNN Dataset name
not specified
100,000
images
Acc =
87.84%
AUC Score
= 94.6%
Public dataset
Mohamed et al.
[20]
Detection of
BC with the
help of
infrared
technology
CNN
ResNet18
GoogleNet
AlexNet
VGG16
Proposed CNN
(DMR-IR)
1000 frontal
thermogram
images
Acc =
93.3%
79.33%
50.0%
100%
99.33%
Need to look
at DL models
that can use
thermal
pictures to
highlight and
name fault
regions
Gogoi et al. [17] Early
abnormality
detection in
breast using
Infrared Breast
Thermography
technique
SVM—support
vector machine
Dataset of
breast
thermogram
images from
60 females
Acc =
83.22%
Inclusion of
dataset of
asymptomatic
patients.
Expansion of
dataset
Rajasekaran
Subramanian
et al. [18]
To detect and
classify BC
lesions in
mammograms
automatically
Faster R-CNN MIAS 330
mammography
images
mAP (mean
average
precision)
value =
0.857
Future work
to include
CBIS-DDSM
database
Agarwal et al.
[21]
Detection of
tumor in
FFDM
mammogram
images
Faster R-CNN OPTIMAM
Mammography
Image
Database
(OMI-DB)
~80,000
FFDMs
True
positive rate
(TPR) =
0.93
–
Remote Breast Cancer Patient Monitoring System … 127
6 Conclusion
The concept of smart healthcare is examined in this review paper, and it is explained
how remote patient monitoring is becoming useful in every element of healthcare.
A general trend of shifting from conventional to smart healthcare is discussed in the
paper. Various breast cancer detection screening approaches and prominent public
databases are highlighted. In the study, major issues in remote monitoring system
are discussed. A case study on remote monitoring system for breast cancer patients
to provide enhanced solution for women in rural areas is also presented in this paper.
Future study will include the development of a feasible system solution for remote
breast cancer patient monitoring that may be employed in primary health clinics.
References
1. Kumar Y, Gupta S, Singla R et al (2021) A systematic review of artificial intelligence techniques
in cancer prediction and diagnosis. Archi Comput Methods Eng State Art Rev 1–28
2. Iranpak S, Shahbahrami A, Shakeri H (2021) Remote patient monitoring and classifying using
the Internet of Things platform combined with cloud computing. J Big Data 8(Article number
120) 1–22
3. Baker SB, Xiang W, Atkinson I (2017) IoT for smart healthcare: technologies, challenges and
opportunities. IEEE Access 5:26521–26544
4. Liaqat M, Iqbal MJ, Javed Z et al (2021) Clinical applications of artificial intelligence
and machine learning in cancer diagnosis: looking into the future. Cancer Cell Int 21(Article
number 270)
5. Malasinhe LP, Ramzan N, Dahal K (2019) Remote patient monitoring: a comprehensive study.
J Ambient Intell Humaniz Comput 10:57–76
6. El-Rashidy N, El-Sappagh S, Riazul Islam SM (2017) Mobile health in remote patient
monitoring for chronic diseases: principles, trends, and challenges. Diagnostics 11(4)
7. Qadri YA, Nauman A, Zikria YB et al (2020) The future of healthcare IoT: a survey of emerging
technologies. IEEE Comm Surveys Tutori 22(2):1121–1167
8. Alanazi SA, Kamruzzaman MM, Nazirul Md et al (2021) Boosting breast cancer detection
using convolutional neural network. J Healthcare Eng 2021(Article ID 5528622), 11 p
9. Vakaa AR, Sonia B, Sudheer Reddy K (2020) Breast cancer detection by leveraging machine
learning. Science Direct ICT Express 6(4):320–324. ISSN 2405-9595
10. Taleb H, Nasser A, Andrieux G et al (2021) Wireless technologies, medical applications and
future challenges in WBAN: a survey. Wirel Netw 27:5271–5295
11. Salvi S, Kadam A (2021) Breast cancer detection using deep learning and IoT technologies.
Journal of Physics: Conference Series 1831 012030. In: International conference on robotics
and artificial intelligence (RoAI)
12. Abdelhafiz D, Yang C, Ammar R, Nabavi S (2019) Deep convolutional neural networks for
mammography: advances, challenges and applications. Bio Inform 20(Article number 281)
13. Mashekova A, Zhao Y, Ng EYK et al (2022) Early detection of the breast cancer using infrared
technology—A comprehensive review. In: Thermal science and engineering progress, vol 27.
Elsevier
14. Mohamed EA, Rashed EA, Gaber T, Karam O (2022) Deep learning model for fully automated
breast cancer detection system from thermograms. PloS Public Library Sci One 17(1):e0262349
15. Rehman O, Farrukh Z, Al-Busaidi AM et al (2020) IoT powered cancer observation system.
SMA 2020, pp 313–318
128 S. Parshionikar and D. Bhattacharyya
16. Halim A, Andrew A, Melvin A et al (2019) Existing and emerging breast cancer detection
technologies and its challenges: a review. Appl Sc 11(22):10753
17. Gogoi UR, Majumdar G, Bhowmik MK. Evaluating the efficiency of infrared breast thermog-
raphy for early breast cancer risk prediction in asymptomatic population. Infrared Phys Technol
99,:201–211
18. Rajasekaran Subramanian D, Rubi D, Lakshmi RG et al (2020) Breast cancer lesion detection
and classification in radiology images using deep learning. Eur J Mol Clin Med 07(3)
19. Hamim M, Paul S, Hoque SI et al (2019) IoT based remote health monitoring system for patients
and elderly people. In: International conference on robotics, electrical and signal processing
techniques (ICREST). INSPEC Accession Number 18473342
20. Pradhan B, Bhattacharyya S, Pal K (2021) IoT-based applications in healthcare devices. J
Healthcare Eng 2021 (article ID 6632599) 18 p
21. Agarwal R, Díaz O, Yap MH et al (2020) Deep learning for mass detection in full field digital
mammograms. Comput Biol Med 121:103774. ISSN 0010-4825
Simplifying the Code Editor Using
MEAN Stack Technologies
S. NagaMallik Raj , M. Jyothsna, P. Srinu, S. Karthik, K. Gnana Jeevana,
N. Thirupathi Rao , and Debnath Bhattacharyya
Abstract The online coding platform lies on the remote server which can be
accessed through browsers. As we know that online code editors are efficient and fast,
it is a popular tool among developers. Competitive coding is an essential skill that
should be possessed by every graduate. Though we have various coding platforms
over the Internet, having our coding platform for our college brings and develops the
competitive environment in our college. This platform enables students to practice
coding questions by facilitating multiple programming languages. They can compete
with their fellow parallel students which keeps the competitive environment. Here,
senior and pro students can become problem setters and every individual will be given
a chance to contribute their questions or blogs or articles. Here, end application will
be a web application that can be accessed by every user upon authentication. This
platform makes analytics and prepares reports about the user performance where
they are good and where they must improve based on previously solved questions.
Keywords Code editor ·MEAN stack ·Node.js
1 Introduction
Nowadays, information and communication technology (ICT) plays an important
role in enhancing quality and support in the engineering-pedagogical system. By
constructing computer laboratories with Internet access, the government of India has
S. NagaMallik Raj (B) · N. T. Rao
Department of Computer Science & Engineering, Vignan’s Institute of Information Technology
(A), Visakhapatnam, AP, India
e-mail: mallikblue@gmail.com
M. Jyothsna · P. Sr i n u · S. Karthik · K. G. Jeevana
Department of CSE, Vignan’s Institute of Information Technology (A), Duvvada, Visakhapatnam,
India
D. Bhattacharyya
Department of Computer Science & Engineering, Koneru Lakshmaiah Education, Vaddeswaram,
Guntur, Andhra Pradesh 522502, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_13
129
130 S. NagaMallik Raj et al.
taken the lead in introducing ICT to so many education levels. In the Indian educa-
tion system, this strategy allows every student to study and enhance his knowledge
without restrictions, while ICT introduces many new teaching–learning processes.
An online coding platform has been regarded as a creative teaching technique that
helps students to tackle coding difficulties virtually from anywhere and at any
moment. Flexibility, portability, low cost, and user-friendliness. Despite various
challenges, such as network issues, online coding platforms have been regarded
as useful tools for acknowledging and enhancing students’ programming abilities.
The faculty’s workload and time will be reduced because of this method [1–4].
There are many coding platforms out there in Internet but none of them are
providing college-specific leaderboards. No platform can create a competitive envi-
ronment in college level. We cannot find any platform where a student or user can
contribute their coding questions, but only admins can.
MEAN is an acronym that stands for MongoDB, Express.js, AngularJS, and
Node.js. MEAN is a complete JavaScript stack that is mostly used for cloud-ready
apps. Understanding why you may use it, identifying examples of when you could
use it, and delving further into the different components can all help you optimize
MEAN’s usefulness for software development [5–8].
1.1 Why MEAN?
The below aspects made us choose MEAN stack.
•Lower development costs
•High performance
•Open-source stack
•User-friendly
•Community support
•Infinite modules
•Flexibility and scalability.
Some major established applications built on MEAN are Facebook, Instagram,
YouTube, Forbes, and Paytm.
As all students cannot upload the questions into the platform, this project helps
students in contributing the questions. We can only answer the questions available
in global platforms, but if we are interested in contributing the questions, it is not
possible for us to do that. So, this project aims in helping the students (Fig. 1).
Simplifying the Code Editor Using MEAN Stack Technologies 131
Fig. 1 Working model of
MEAN stack
2 Related and Proposed Work
2.1 Existing Systems
Based on the review of the existing systems, the below are the observations.
Hacker Rank
Hacker Rank is a well-known coding platform that allows programmers across the
world to solve coding problems and challenges. Hacker Rank holds up a wide range
of programming languages and spans a wide range of platforms. In a wide range of
computer science areas when a coder uses when a programmer submits a solution to a
programming challenge, the correctness of the output is used to score the submission
[4, 8–10].
Code Chef
It is a well-known competitive programming platform which is a global one that
supports over 50 programming languages and has a significant programming commu-
nity that helps students and other computer professionals test and improve their
coding abilities. Its goal is to create a forum for students and professional software
engineers to practice, compete, and grow. Code chef holds frequent practice contests
for ACM-IPCCs as well as monthly contests with rewards [11, 12].
In the existing platforms, only the campus ambassadors and platform admins of
those platforms can contribute a question, but others do not have that facility.
2.2 Proposed System
Our project focuses on helping students in our college. Till now, we do not have any
college-specific coding platform. So, keeping in that mind we thought of providing
support to the students of our college. As we have mentioned before, in the existing
systems students did not have a chance to post a question. Solving a problem and
creating a problem are equally important [13–15]. Problem designing improves a
132 S. NagaMallik Raj et al.
Fig. 2 An overview of proposed system
student’s thinking capability and skill. So, our project facilitates that feature where
users can contribute the questions. To do that, admin validation is necessary. Once the
admin approves the question contributed by a student. That question can be directed
to the question list which will be visible on the home page. Now, coming to the code
editor, it is being supported by multiple programming languages such as C, C++,
and Python. Various varieties of themes are also provided. It is not just a code editor;
test cases also will be given with the question itself. Once the user submits the code,
he can check out the results and work on that if needed. The other important feature
our project consists of is blog creation. We can also write blogs on this platform. The
person who has created the blog and when it was done will be displayed at the bottom
of the blog name. Users can see his blogs by going into his profile and clicking my
blogs [16].
This is a college-specific platform with some new features which the existing
system lacks (Fig. 2).
3 Results and Sample OUTPUT Screens
Below are the sample screenshots for our proposed system. Firstly, user has to register
into the proposed platform using the registration page in Fig. 3. So, after user regis-
tration, they will be directed to a login page, which needs a user to enter certain
inputs such as his ID number and password as shown in Fig. 4. Then, the user will be
allowed into the home page, which lists a set of questions along with their difficulty
levels as shown in Fig. 5. This is what the code editor looks like which was shown in
Fig. 6. The page will contain a question description on the left side and an editor on
the right side for the user. Figure 7 shows how results will be shown after submitting
the solution along with the time and memory taken for each input.
Simplifying the Code Editor Using MEAN Stack Technologies 133
Fig. 3 User registration
page
Fig. 4 Login page for the
user
134 S. NagaMallik Raj et al.
Fig. 5 Home page of
proposed model
Fig. 6 Sample screen for
code editor
Fig. 7 Proposed model with
questionnaire part and editor
part
In our proposed model, we have added additional features like blogs and a question
contribution page which are shown in Figs. 8 and 9. Figure 10 depicts the blog page,
i.e., whatever blogs are created by users will be displayed here.
Simplifying the Code Editor Using MEAN Stack Technologies 135
Fig. 8 Sample screen for result area
Fig. 9 Question contribution page
Fig. 10 Blogs that users
have contributed will be
displayed here
136 S. NagaMallik Raj et al.
4 Conclusion
Hereby, we conclude that this project is going to be a beneficial platform for the
students. We worked up the features that are missing in the existing system, and we
provided our system with those features. To prepare for campus placements specially
to gain or brush up on coding skills, we do not need to think of other platforms. As
this is a college-specific platform, students can make use of this amazing opportu-
nity. Every student can check his ranking in the college. This project will empower
competitive spirit in students which is a vital task. Creating a question is not such an
easy task because it needs a problem setter to think of every constraint. Test cases
will be formed accordingly. Problem setter can be any student. It does not matter
whether the person is senior or junior. Students cannot contribute the question just
like that. Once he prepares the question, he needs to wait until the admin’s approval.
References
1. Satyanarayana KV, Rao NT, Bhattacharyya D, Hu Y (2022) Identifying the presence of
bacteria on digital images by using asymmetric distribution with k-means clustering algorithm.
Multidimension Syst Signal Process 33(2):301–326. https://doi.org/10.1007/s11045-021-008
00-0
2. Chandra Sekhar P, Thirupathi Rao N, Bhattacharyya D, Kim T (2021) Segmentation of natural
images with k-means and hierarchical algorithm based on mixture of Pearson distributions. J
Sci Ind Res 80(8):707–715. Retrieved from www.scopus.com
3. Bhattacharyya D, Dinesh Reddy B, Kumari NMJ, Rao NT (2021) Comprehensive analysis on
comparison of machine learning and deep learning applications on cardiac arrest. J Med Pharm
Allied Sci 10(4):3125–3131. https://doi.org/10.22270/jmpas.V10I4.1395
4. Bhattacharyya D, Doppala BP, Thirupathi Rao N (2020) Prediction and forecasting of persistent
kidney problems using machine learning algorithms. Int J Curr Res Rev 12(20):134–139.
https://doi.org/10.31782/IJCRR.2020.122031
5. Mandhala VN, Bhattacharyya D, Vamsi B, Thirupathi Rao N (2020) Object detection using
machine learning for visually impaired people. Int J Curr Res Rev 12(20):157–167. https://doi.
org/10.31782/IJCRR.2020.122032
6. Bhattacharyya D, Kumari NMJ, Joshua ESN, Rao NT (2020) Advanced empirical studies on
group governance of the novel corona virus, mers, sars and ebola: a systematic study. Int J Curr
Res Rev 12(18):35–41. https://doi.org/10.31782/IJCRR.2020.121828
7. Asish Vardhan K, Thirupathi Rao N, Naga Mallik Raj S, Sudeepthi G, Divya, Bhattacharyya
D, Kim T (2019) Health advisory system using IoT technology. Int J Recent Technol Eng
7(6):183–187. Retrieved from www.scopus.com
8. Eali SNJ, Bhattacharyya D, Nakka TR, Hong S (2022) A novel approach in bio-medical image
segmentation for analyzing brain cancer images with U-NET semantic segmentation and TPLD
models using SVM. Traitement Du Signal 39(2):419–430. https://doi.org/10.18280/ts.390203
9. Doppala BP, NagaMallik Raj S, Stephen Neal Joshua E, Thirupathi Rao N (2021) Automatic
determination of harassment in social network using machine learning. https://doi.org/10.1007/
978-981-16-1773-7_20. Retrieved from www.scopus.com
10. Eali SNJ, Rao NT, Swathi K, Satyanarayana KV, Bhattacharyya D, Kim T (2018) Simulated
studies on the performance of intelligent transportation system using vehicular networks. Int J
Grid Distrib Comput 11(4):27–36. https://doi.org/10.14257/ijgdc.2018.11.4.03
Simplifying the Code Editor Using MEAN Stack Technologies 137
11. Joshua ESN, Battacharyya D, Doppala BP, Chakkravarthy M (2022) Extensive statistical anal-
ysis on novel coronavirus: towards worldwide health using apache spark. https://doi.org/10.
1007/978-3-030-72752-9_8. Retrieved from www.scopus.com
12. Joshua ESN, Bhattacharyya D, Chakkravarthy M (2021) Lung nodule semantic segmentation
with bi-direction features using U-INET. J Med Pharm Allied Sci 10(5):3494–3499. https://
doi.org/10.22270/jmpas.V10I5.1454
13. Joshua ESN, Bhattacharyya D, Chakkravarthy M, Kim H (2021) Lung cancer classification
using squeeze and excitation convolutional neural networks with grad cam++ class activation
function. Traitement Du Signal 38(4):1103–1112. https://doi.org/10.18280/ts.380421
14. Joshua ESN, Chakkravarthy M, Bhattacharyya D (2021) Lung cancer detection using impro-
vised grad-cam++ with 3D CNN class activation. https://doi.org/10.1007/978-981-16-177
3-7_5 Retrieved from www.scopus.com
15. Neal Joshua ES, Bhattacharyya D, Chakkravarthy M, Byun Y (2021) 3D CNN with visual
insights for early detection of lung cancer using gradient-weighted class activation. J Healthc
Eng. https://doi.org/10.1155/2021/6695518
16. Neal Joshua ES, Chakkravarthy M, Bhattacharyya D (2020) An extensive review on lung cancer
detection using machine learning techniques: a systematic study. Rev d’Intelligence Artificielle
34(3):351–359. https://doi.org/10.18280/ria.340314
Prediction and Identification of Diseases
to the Crops Using Machine Learning
S. NagaMallik Raj , Pyla Lohit, Doddala Jyo-theendra,
Kannuru Chandana, P. Nikhil, N. Thirupathi Rao ,
and Debnath Bhattacharyya
Abstract Farming is one of the major sectors that influence a country’s economic
growth. In countries like India, majority of the population is depending on agriculture
for their livelihood. But in recent times, agriculture in India is enduring a structural
change leading to a disastrous situation. The main purpose of this project is building
a website to assist people who wants to grow crops at their home or interested
in terrace farming and assist the farmers to maximize their yield and to sell the
harvested crop online by themselves without involving any middleman in between
them so that the farmer will enjoy maximum possible profits solely by himself. Our
website reduces time and effort of users by providing various applications such as
crop recommendation which works by analyzing various attributes such as location,
amount of rainfall in the region, and soil pH values; fertilizer recommendation which
recommends the necessary organic measures to be taken based on the type of crop
and soil NPK values; and crop disease prediction which works in order to predict the
disease of a particular crop by uploading the crop image and suggests the organic
treatment for that particular crop accordingly. Farmer can choose the type of treatment
for their crops.
Keywords Farmers ·Middlemen ·Agriculture ·Rainfall ·Crop
recommendation ·Fertilizer recommendation ·Terrace farming ·Crop disease
prediction ·Soil pH values
S. NagaMallik Raj (B) · N. T. Rao
Department of Computer Science & Engineering, Vignan’s Institute of Information Technology
(A), Visakhapatnam, AP, India
e-mail: mallikblue@gmail.com
P. Lohit · D. Jyo-theendra · K. Chandana · P. Nikhil
Department of CSE, Vignan’s Institute of Information Technology (A), Duvvada, Visakhapatnam,
India
D. Bhattacharyya
Department of Computer Science & Engineering, Koneru Lakshmaiah Education, Vaddeswaram,
Guntur, Andhra Pradesh 522502, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_14
139
140 S. NagaMallik et al.
1 Introduction
Agriculture and farming are one of the main sectors that influence country’s economy.
Farming (irrespective of scale) requires keeping learning new information regarding
the farming techniques to make sure they get proper yield and profits from what they
have cultivated. A friendly agriculture guidance and knowledge gathering service can
be very resourceful and helpful for the farmers. Crop cultivation and selling depend
upon several factors which include the climate, average rainfall in that area, and
several other factors. This application helps farmers to sell their products online and
leading them to achieve success and increase in their standard of living. Agriculture
has always been an intensive record in India. India is ranked second in the farm
output global [1–5]. Percentage Share of GVA of Agriculture and Allied sector
to Indian Total Economy is at 20% in the year 2020–2021 despite the covid-19
pandemic [6–8]. The crop yield relies upon on multiple factors consisting of soil,
climatic geographic, organic, and economic elements. These critical factors should
be considered by farmers for selection of crop. These factors play major role in crop
yield. It is hard for farmers to determine while and which plants to plant due to
fluctuating market charges. Bringing up to Wikipedia figures, India’s suicide rate
has ranged from 1.4–1.8% in line with 100,000 populations, during the last 10 years
[4, 8, 9]. It is tough for farmers to determine while and which crops to plant and
what is the right time and place to start due to uncertainty in climatic conditions.
Using various fertilizers is likewise unsure because of changes in seasonal climatic
conditions and fundamental belongings such as soil, water, and air. In this situation,
the crop yield rate is steadily declining. The solution to the hassle is to provide
a smart user-friendly recommendation system to the farmers. And there are many
diseases that are destroying the plants. The disease like fungi and bacteria lives on
the plant and takes their energy from them. And, with excessive use of pesticides
and insecticides also they are suffering from diseases. These diseases are responsible
for great deal of damage for farmers. By identifying the diseases, the rotten parts of
the plant, the crops can be saved in the early stage. This project aims to improve the
benefits and profits of the farmers.
2 Literature Review
Farming has been the primary occupation in India. Farming has become the major
livelihood for farmers. The farmers in India are not getting proper benefits as they
sell their crops to the dealers in the nearest market. The main agenda of this project is
to maximize the profits enjoyed by the farmers. In this project, the farmers will get an
interface where they can perform advertising of their marketing, get the current rates
of market, and get in touch with the clients through this website directly. They can
sell their products directly to the clients. The users can access this website with an
active Internet connection. The user can access recommendation systems from this
Prediction and Identification of Diseases to the Crops … 141
website. The applications are provided on this website to guide the farmers. The user
can get to know what crop and fertilizers to use. With the help of machine learning,
the crop and the fertilizer recommendation can be done. The user can access it by
giving the soil values, location, and crop as input. This project comes with disease
prediction where the user has to upload the image of a diseased plant.
3 Motivation
There exists a social responsibility to make sure that the farmers are getting the
maximum profit for all the hard work without any interference from middlemen in
between. Our nation’s past, present, and future cannot be understood without farmers
because the roots of our country are deeply connected with agriculture. Our website
serves as a one-stop solution for farmers and homestead farmers by recommending
the best crops and fertilizers based on soil test results, and by simply uploading an
image, it identifies the disease that affected the plant. Additionally, the farmer can
sell his harvested crop by himself via a simple interface that connects him directly to
the public. This educates and motivates the people toward organic farming leading
to a healthier lifestyle.
4 Proposed System
The proposed system can be accessed through a mobile or a desktop with an Internet
connection. This system is solely dedicated to the farmers and the people who want
to farm. This website allows the users to better understand and gather necessary
information regarding the crops they want to cultivate and also how to treat them.
This proposed system lay outs a plan to sell the cultivated crops in bulk without
any interference from middlemen which allows maximizing the profits of the user.
Figure 1 represents the home page of the website, Fig. 2 represents the disease
prediction page, and Fig. 3 represents the services view page.
5Flowchart
See Fig. 4.
142 S. NagaMallik et al.
Fig. 1 Implementation of
the proposed model showing
the home page
Fig. 2 Implementation of
the proposed model showing
the modules
Fig. 3 Implementation of
the proposed model showing
abstract view
6 Algorithms, Dataset Used, and Accuracy Comparison
The crop recommendation model was developed using r andom forest algorithm after
testing the accuracy of the models using different algorithms such as decision tree
algorithm, random forest algorithm, SVM classifier algorithm, and Naive Bayes algo-
rithm. Below image Fig. 5 represents an assessment of the accuracy of the different
algorithms including decision tree, random forest, SVM classifier, and Naïve Bayes
for the crop recommendation model.
Prediction and Identification of Diseases to the Crops … 143
Fig. 4 Flowchart of the
proposed system
Fig. 5 Accuracy
comparison between
different algorithms
Precision agriculture is in trend nowadays. It helps the farmers to get informed
decision about the farming strategy. Hence, we used a dataset to build a predictive
model to recommend the most suitable crops to grow in a particular farm based on
various parameters.
The data used in this project is made by augmenting and combining various
publicly available datasets of India like weather, soil, etc. This data is relatively
simple with very few but useful features unlike the complicated features affecting
the yield of the crop.
The data has Nitrogen, Phosphorous, Potassium, and pH values of the soil. Also,
it contains the humidity, temperature, and rainfall required for a particular crop.
The data used in this project for fertilizer suggestion is custom built with appro-
priate NPK values corresponding to a particular crop; hence, when the values are
144 S. NagaMallik et al.
entered in the application, it will predict what the soil lacks or has excess of and will
recommend improvements.
The plant disease prediction is done using the classification with the help of
RESNET-9 which resulted in 99.2% accuracy for the used dataset, and the dataset
is created using offline augmentation from the original dataset. The original Plant
Village Dataset can be found here [10–13]. This dataset consists of about 87 K rob
images of healthy and diseased crop leaves which is categorized into 38 different
classes. The total dataset is divided into 80/20 ratio of training and validation set
preserving the directory structure. A new directory containing 33 test images is
created later for prediction purpose.
7 Conclusion
This work will be helpful for farmers and anyone willing to farm to know to get
more information about their crops and issues affecting their yield right from their
homes. The prediction of crop yield is primarily based on soil records and a few
other factors which affect crop yield. From the above work, we conclude that the
following functions can be achieved by building an Internet website for farmers. It
helps farmers by uploading a photo of the crop. Crop disease detection makes use of
image processing wherein users get insecticides primarily based on disease photos.
This efficient way of visualizing the crop diseases in plants saves cost by avoiding
the unnecessary usage of the fertilizers, insecticides, and pesticides, and the fertil-
izer prediction and crop prediction can be done based on soil conditions [14–16].
This project additionally says what fertilizers must be used instead of depending
on the farmer’s earlier experience. The project has successfully carried out the idea
of an online enterprise and provided the e-commerce application to sell their prod-
ucts online so the farmers can be benefited from the whole profit themselves. The
buyers and dealers can get the information required to go and purchase the crop yield
from anywhere. This way the customers can know who produced the crops they are
buying. This website helps people with a lack of insights on farming. This system is
successfully developed with various applications all in one place for farmers.
References
1. Satyanarayana KV, Rao NT, Bhattacharyya D, Hu Y (2022). Identifying the presence of
bacteria on digital images by using asymmetric distribution with k-means clustering algorithm.
Multidimension Syst Signal Process 33(2):301–326. https://doi.org/10.1007/s11045-021-008
00-0
2. Chandra Sekhar P, Thirupathi Rao N, Bhattacharyya D, Kim T (2021) Segmentation of natural
images with k-means and hierarchical algorithm based on mixture of Pearson distributions. J
Sci Ind Res 80(8):707–715. Retrieved from www.scopus.com
Prediction and Identification of Diseases to the Crops … 145
3. Bhattacharyya D, Dinesh Reddy B, Kumari NMJ, Rao NT (2021) Comprehensive analysis on
comparison of machine learning and deep learning applications on cardiac arrest. J Med Pharm
Allied Sci 10(4):3125–3131. https://doi.org/10.22270/jmpas.V10I4.1395
4. Bhattacharyya D, Doppala BP, Thirupathi Rao N (2020) Prediction and forecasting of persistent
kidney problems using machine learning algorithms. Int J Curr Res Rev 12(20):134–139.
https://doi.org/10.31782/IJCRR.2020.122031
5. Mandhala VN, Bhattacharyya D, Vamsi B, Thirupathi Rao N (2020) Object detection using
machine learning for visually impaired people. Int J Curr Res Rev 12(20):157–167. https://doi.
org/10.31782/IJCRR.2020.122032
6. Bhattacharyya D, Kumari NMJ, Joshua ESN, Rao NT (2020) Advanced empirical studies on
group governance of the novel corona virus, mers, sars and ebola: a systematic study. Int J Curr
Res Rev 12(18):35–41. https://doi.org/10.31782/IJCRR.2020.121828
7. Asish Vardhan K, Thirupathi Rao N, Naga Mallik Raj S, Sudeepthi G, Divya, Bhattacharyya
D, Kim T (2019) Health advisory system using IoT technology. Int J Recent Technol Eng
7(6):183–187. Retrieved from www.scopus.com
8. Eali SNJ, Bhattacharyya D, Nakka TR, Hong S (2022) A novel approach in bio-medical image
segmentation for analyzing brain cancer images with U-NET semantic segmentation and TPLD
models using SVM. Traitement Du Signal 39(2):419–430. https://doi.org/10.18280/ts.390203
9. Doppala BP, NagaMallik Raj S, Stephen Neal Joshua E, Thirupathi Rao N (2021) Automatic
determination of harassment in social network using machine learning. https://doi.org/10.1007/
978-981-16-1773-7_20. Retrieved from www.scopus.com
10. Eali SNJ, Rao NT, Swathi K, Satyanarayana KV, Bhattacharyya D, Kim T (2018) Simulated
studies on the performance of intelligent transportation system using vehicular networks. Int J
Grid Distrib Comput 11(4):27–36. https://doi.org/10.14257/ijgdc.2018.11.4.03
11. Joshua ESN, Battacharyya D, Doppala BP, Chakkravarthy M (2022) Extensive statistical anal-
ysis on novel coronavirus: towards worldwide health using apache spark. https://doi.org/10.
1007/978-3-030-72752-9_8. Retrieved from www.scopus.com
12. Joshua ESN, Bhattacharyya D, Chakkravarthy M (2021) Lung nodule semantic segmentation
with bi-direction features using U-INET. J Med Pharm Allied Sci 10(5):3494–3499. https://
doi.org/10.22270/jmpas.V10I5.1454
13. Joshua ESN, Bhattacharyya D, Chakkravarthy M, Kim H (2021) Lung cancer classification
using squeeze and excitation convolutional neural networks with grad cam++ class activation
function. Traitement Du Signal 38(4):1103–1112. https://doi.org/10.18280/ts.380421
14. Joshua ESN, Chakkravarthy M, Bhattacharyya D (2021) Lung cancer detection using impro-
vised grad-cam++ with 3D CNN class activation. https://doi.org/10.1007/978-981-16-177
3-7_5. Retrieved from www.scopus.com
15. Neal Joshua ES, Bhattacharyya D, Chakkravarthy M, Byun Y (2021) 3D CNN with visual
insights for early detection of lung cancer using gradient-weighted class activation. J Healthc
Eng 2021. https://doi.org/10.1155/2021/6695518
16. Neal Joshua ES, Chakkravarthy M, Bhattacharyya D (2020) An extensive review on lung cancer
detection using machine learning techniques: a systematic study. Rev d’Intelligence Artificielle
34(3):351–359. https://doi.org/10.18280/ria.340314
Pulse-Based Smart Electricity Meter
Using Raspberry Pi and MEFN
Eswar Abisheak Tadiparthi, Majji Prasanna Kumari,
Basanaboyana Vamsi Sai, Kollana Bharat Kalyan, B. Dinesh Reddy ,
N. Thirupathi Rao , and Debnath Bhattacharyya
Abstract As the computation power of microcomputers increases over time, we
need to put them to proper use. Green energy generation and consumption are evident
problems that need to be prioritized. We know that green energy is not produced at
a scale for us to rely on but if we know how electricity is consumed, we might be
able to hack our way. In order to know that we are using light-dependant resistor
(LDR) to count pulses from the electricity meter and update it in the database using
a Raspberry Pi connected to the Internet which hits the server, the end user can see
the utilization in real time.
Keywords Smart electricity meter ·Raspberry Pi ·LDR ·Flutter ·MERN
1 Introduction
As humans, we continue to innovate and push boundaries to rely on carbon-free
energy generation and consumption [1]. One needs to understand how humans
consume energy, so one can start switching to carbon-free energy sources as not
all areas are hungry for energy. The electricity that is consumed by the user can be
monitored in real time using Raspberry Pi with an Internet connection; based on
the local energy consumption tariffs, the server can calculate the cost on the day
of their billing cycle meanwhile push notifications to their smartphones and send
emails to keep them updated on their usage. There might be electrical leakage which
is dangerous and causes a higher bill amount for the user. It might be due to malfunc-
tions of electronic devices [2], old devices which are not great at optimal consumption
of electricity, or flowing to the ground through a bad connection. Most appliances that
E. A. Tadiparthi (B) · M. P. Kumari · B. V. Sai · K. B. Kalyan · B. Dinesh Reddy · N. T. Rao
Department of Computer Science and Engineering, Vignan’s Institute of Information Technology,
Visakhapatnam, Andhra-Pradesh, India
e-mail: teswar2001@gmail.com
D. Bhattacharyya
Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation,
Vaddeswaram, Guntur, Andhra-Pradesh 522502, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_15
147
148 E. A. Tadiparthi et al.
were designed and built recently are optimized to consume less energy to operate.
So, both the producer and consumer can be benefitted.
2 Literature
There are many papers published on the concept of smart electricity meter or auto-
matic meter reading. Most of the implementations use GSM, current, and voltage
sensing units. This paper focuses on using FCM (Firebase Cloud Messaging) to send
notifications and node mailer to send emails, and using the proposed method, a single
Raspberry Pi could read and sync meter readings from multiple meters simultane-
ously, and reading from pulses/blinks is more accurate than the reading from sensing
units or the LCD display [3].
Reference [4] Due to manual labor, the current electricity billing system has
major problems. This system will provide meter read and power disconnection when
power consumption exceeds the stated limit using IoT. The Arduino esp8266 micro-
controller is designed to perform objectives with the help of the GSM module. It is
proposed that all existing energy meter problems be overcome. All information is
sent to the consumer cell phone via IoT and GSM module and is also displayed on
the LCD. It saves time and helps to eliminate human interference using IoT.
3 Proposed Methodology
A system would continuously monitor and update the electricity usage and keep the
user in the loop and send timely alerts to caution and alert the user so one can take
necessary action to optimistically consume non-renewable resources which would
save money and help the environment. Electricity meters have a blinking/flashing
LED, often with small text that reads 1000 Imp/kWh. The two important things here
are that you have a blinking LED and you know the number of impulses that results in
1 unit, e.g., 800. LDR senses the blinks [5] and makes a get request to the server with
the user id; the server would calculate the consumption from the no blinks/pulses
which is converted into units (kWh) based on meter specifications. The data can be
used to do further analysis as it is granular.
4 System Design
Three main connected systems would help achieve this [6], and a puller at the meter
with an Internet connection would poll every pulse. The puller will require a server
that would store and do further analysis. The user will require an interface to view
Pulse-Based Smart Electricity Meter Using Raspberry Pi and MEFN 149
and receive updates. The data stored can be used to do intensive analysis in almost
real time.
In Figs. 1 and 2, the server is written in NodeJS where the routing and middleware
are handled using express and JWT (jsonwebtoken) for authentication, and it can be
hosted in any cloud service provider such as GCP, AWS, or Heroku. As we do not
have much computation. The data is stored as pulsed in cloud-based MongoDB
which is a document-oriented NoSQL database. The database has three types of
objects FCM, pulse, and user. A single user can have multiple FCM where he can
log in to multiple devices with the same account A pulse is just a timestamp of when
the blink happened with a unique id. We compute the units consumed based on the
no of pulses so there is a field that maintains the count of pulses. To connect LDR
to RPI, we use GPIO. It is a standard interface used to connect microcontrollers to
other electronic devices. In this case, we are using those pins to connect multiple
LDR to a single Raspberry Pi.
The client app is built using flutter which has the following screens sign-in, signup,
usage, profile, and pulse list. Flutter is an open-source front-end framework for
creating native mobile applications and supports cross-platform development with a
single codebase. Developing front-end applications in flutter is fast and flexible, and
it has a huge community of developers to have support was shown in Fig. 3.
Fig.1 System design
150 E. A. Tadiparthi et al.
Fig. 2 Connection diagram
5 Conclusion
From the hourly breakdown of the power consumption for each day, one can under-
stand which appliances are consuming the most energy and one could optimally use
that appliance to reduce the consumption. Overall consumption data can be used to
analyze further and suggest areas where we can shift to rely on carbon-free electricity
Pulse-Based Smart Electricity Meter Using Raspberry Pi and MEFN 151
Fig. 3 Showing the client
and server architecture app
using flutter
(a) (b)
(c) (d)
consumption without any problem. One can also find out any electricity leaks and
malfunctioning electronic devices. Electricity is the most important resource in daily
life and is important for everyone to not waste it, was shown in Fig. 4.
This method would also help limit the need for humans to repeatedly check usage
at the meter. It is very cost effective when this is implemented over time but initially
one has to invest to buy and set up, and later one only needs to maintain them and have
few support members to fix any issues. Raspberry Pi is very efficient in performance
for its power consumption. A single RPI can support multiple LDR which can vary
depending upon the model and available GPIO pins, so it would be cheaper to read
from multiple meters.
152 E. A. Tadiparthi et al.
Fig. 4 Annual cost versus benefits
References
1. Zheng G, Zhang Z (2011) Intelligent wireless electric power management and control system
based on ZigBee technology. In: International conference on transportation, mechanical, and
electrical engineering (TMEE) Changchun, China, 16–18 Dec 2011, pp 1120–1124
2. Maitra S (2008) Embedded energy meter—A new concept to measure the energy consumed by
a consumer and to pay the bill. In: Power system technology and IEEE Power India conference,
2008
3. Monteiro K, Rocha É, Silva É, Santos GL, Santos W, Endo PT (2018) Developing an e-health
system based on IoT Fog and cloud computing. In: 2018 IEEE/ACM International conference
on utility and cloud computing companion (UCC Companion), pp 17–18
4. Popa M (2011) Gateway design and implementation in an automatic meter reading system based
on power line communication. In: 7th International conference on networked computing and
advanced information management (NCM), pp 295–298
5. Mandhala VN, Bhattacharyya D, Vamsi B, Thirupathi Rao N (2020) Object detection using
machine learning for visually impaired people. Int J Curr Res Rev 12(20):157–167. https://doi.
org/10.31782/IJCRR.2020.122032
6. Patel K, Patel SM (2016) Internet of Things-IOT: definition characteristics architecture enabling
technologies application & future challenges. IJESC 6:6122–6131
Brain Tumor Segmentation Using U-Net
Paturi Jyothsna, Mamidi Sai Sri Venkata Spandhana, Rayi Jayasri,
Nirujogi Venkata Sai Sandeep, K. Swathi , N. Marline Joys Kumari ,
N. Thirupathi Rao , and Debnath Bhattacharyya
Abstract Brain tumor segmenting from the non-invasive magnetic resonance
imaging (MRI) is hard and the most vital task for several applications in the area
of medical science analysis.In current days, surgical operations are usually done
on hand-operated ways in the hospital that takes excess time. Manually segmenting
the brain tumor is really a very overlong job, and it much more depends on the
individual person, and we found that gliomas are the hardest tumor to be found out
having irregular shape and vague boundaries. MRI images are the mostly used for the
segmentation of the brain affected portion. Segmentation method for MRI images of
brain is one of the ways that radioscopy performs on the brain image for finding the
tumor tissue from the normal tissue. In this paper, we present this proposed approach
depending on fully convolutional network (FCN) and we are making use of U-Net
as the model. This model can be used as a vital essential on prearranged surgical
operations to accomplish the successful operations of human brain.
Keywords Brain tumor segmentation ·Magnetic resonance image ·Fully
convolutional network ·Deep learning ·Machine learning
P. Jyothsna (B) · M. S. S. V. Spandhana · R. Jayasri · N. V. S. Sandeep · K. Swathi · N. T. Rao
Department of Computer Science & Engineering, Vignan’s Institute of Information Technology
(A), Visakhapatnam, AP, India
e-mail: paturijyothsna@gmail.com
N. Marline Joys Kumari
Department of Computer Science & Engineering, Anil Neerukonda Institute of Technology and
Sciences, Sanghivalasa, Visakhapatnam, AP 531162, India
D. Bhattacharyya
Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation,
Vaddeswaram, Guntur, AP 522302, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_16
153
154 P. Jyothsna et al.
1 Introduction
Machine learning has been applied in many different areas like agriculture, medical
for the purpose of detection, classification, prediction, and the segmentation. The
benchmark in brain tumor prognosis is biopsied which includes clinical examination
using various anatomical techniques [1]. The process of biopsy is a kind of surgical
activity which results in bleeding and even cause injury that may results in loss of
functioning. In order to get rid of this invasive method, we come up by using the
magnetic resonance images to characterize the brain tumor tissues from the normal
tissues by looking at the MRI images. The analysis of MRI images of brain is a
time-taking task, and these can be attained by the professional radiologists. But in
few cases, even the experienced radiologists cannot find out the tumors. The main
challenging situation in finding the tumor is because of its different size, shape,
boundaries [2], and at different locations. In this paper, we mainly focus on gliomas
tumor. As we know, gliomas tumor is hard to find because of its ambiguousness.
1.1 Brain Tumor Modalities
Noninvasive magnetic resonance imaging is usually used for the purpose of analysis
of brain tumors. We have many non-invasive methods such as magnetic resonance
imaging (MRI) and also other invasive methods to give accuracy to the brain tumor
structure. Using other invasive systems other than MRI is extortionate. The modalities
are as follows: T1-weighted (T1) MRI, T1-weighted with contrast enhancement (T1c)
MRI, T2-weighted (T2) MRI, and the T2-weighted with fluid attenuated inversion
recovery (T2-Flair) MRI. The recognition of all the four modalities is shown in Table
1. The diagrammatic representation of the modalities is represented in Fig. 1.
Table 1 Recognition of imaging modalities
Tissue T1 T1c T2 Flair
CSF Dark Dark Bright Dark
White matter Light Light gray Dark gray Dark gray
Cortex Gray Dark gray Light gray Light gray
Fat Bright Light Light Light
Inflammation Dark Light Bright Bright
Brain Tumor Segmentation Using U-Net 155
Fig. 1 Diagrammatic representation of imaging modalities
2 Literature Review
Reference [3] used a 2D U-Net to segment each 3D MRI volume in slices. This
method is faster to train and test and has less computational requirements, but is
significantly overloaded with parameters (about 35 million parameters) and does not
use 3D contextual information. Reference [4] proposed the use of histogram equal-
ization (HE) and Fuzzy Support Vector Machine (FVSM) classification algorithms to
detect brain tumors. The apprehensive parts from the images were segmented largely
using the MRF method for segmentation approach after the brain MRI was prepro-
cessed with histogram equalization. The MRF method improved tumor segmenta-
tion accuracy, advancing the overall performance of the proposed method. For MRI
brain imaging, Natarajan et al. [5] suggested a brain t umor detection approach. The
MRI brain pictures are first preprocessed using a median filter, after which they
are segmented using threshold segmentation and morphological procedures, and
ultimately, the tumor region is determined using image subtraction. This method
accurately depicts the tumor shape in an MRI brain image. Based on Support Vector
Machines, [3] proposed a hybrid technique for brain tumor detection in MRI images.
The texture and intensity characteristics are applied. A technique for detecting and
classifying brain tumors was proposed by [6]. This method uses segmentation to
extract tumors, GLCM to extract features, and then BPNN and KNN classifiers to
classify the MRI brain images as normal or abnormal. In this paper, we present
a unique U-Net-based 3D fully convoluted [7] segmentation network. A complete
data augmentation strategy was applied in this study to improve the segmentation
accuracy.
3 Materials and Methods
The steps involved in this methodology are gathering the dataset, cleaning the data,
data preprocessing, building the model, training the model, and evaluating the model.
Image dataset: The dataset used in this paper consists of trainable data and data
required for the validation. The name of the dataset is Brats2020 [8] from the MICCAI
156 P. Jyothsna et al.
[9]. The Brats have been focusing on the evaluation of contemporary methods of
intracranial tumor in complicated magnetic resonance image scans. Data prepro-
cessing: In the data preprocessing step, we usually make the noise removal, cleaning
the data, and data augmentation. Cleaning the data: As we know, cleaning the data
is the most important thing in any machine learning project. So, we are cleaning the
data before we are feeding that data into the machine learning model. Data augmen-
tation [10, 11]: We use the data generators to create a batch of images which helps
to make training faster. Because dealing with the large dataset is really a tedious job,
we generate data into various batches. The methodology.
1. Building the model: In this paper, we are using the U-Net architecture which
is a complete fully convolutional. We will build the U-Net model by adding
convolutional layers and max pooling layers. We apply up-sampling and down-
sampling in this model and also make use of s kip connections which will help us
to recover the lost information or features. This model will be mainly used for
biomedical images and make the segmentation of the affected region.
2. Training the model: We use the training data to train the model we use in this
proposed system. The model will get trained with different multi-model images
like flair, T1c, T1, and T2.
3. Splitting the data: We will split the data available for the training into training
data and the testing data. So, during the model evaluation we can test the model
against the testing data.
4. Evaluating the model: We finally evaluate the trained model using the testing
dataset.
3.1 U-Net Architecture
The U-Net model is a part of the convolutional neural network but it expanded with
few new features that have been made to the CNN architecture. It was the very first
segmentation model to deal with biomedical images. This model is used not only to
detect whether the infection is present or not but also to localize the spot by masking
its infection. This architecture is called as U-Net because it is in U-shaped. The
complete architecture has 23 layers in total. This model consists of two sections:
contraction path and the expansion path. The contraction path lies on the left side of
the architecture and is also called as a encoder path. And the expansion path is on
the right side of the architecture and called to be as decoder path. The U-Net model
is shown in underneath Fig. 2 architectures of U-Net.
3.2 Fully Convolutional Network
A fully convolutional network uses convolutional neural networks for transformation
of image pixels into pixel classes. This architecture is mainly used for semantic
Brain Tumor Segmentation Using U-Net 157
Fig. 2 Architectures of U-Net
Fig. 3 Fully convolutional network example
segmentation. They deploy layers such as convolutional layer, max pooling layers,
and up-sampling but it does not use dense l ayers unlike CNN, and this is making
the training much faster. The FCN will get back the feature maps to that of the
input image and that can be achieved by applying the transpose convolutional layer.
This leads to correspondence of output and input image in pixel level. The fully
convolutional network is illustrated in Fig. 3.
4 Results and Discussions
In this paper, after creating the U-Net model we trained the model using training data,
and after that, we test the model using the testing data. After processing the data and
evaluated the input using our model and predicted the output. The underneath figure
shows the original flair image and ground truth by applying model.
158 P. Jyothsna et al.
In Fig. 4, the initial picture is the original image and other one is the ground truth
image of a tumor region. The red colored part is a full tumor. And the other regions
are tumor core, and enhancing tumor is represented using other colors.
The ground truth is mostly used in statistics and machine learning to check the
results of machine learning model for the accuracy. Here, the ground truth r efers the
information obtained during the training.
Figure 5 is an example t o show how one of the sub-regions looks a like by using
segmentation. And Table 2 is describing the Dice coefficient.
Figure 6 illustrates about the representation of the graphs that compares the
training and validation of accuracy, loss function, Dice coefficient, and the mean
IOU. This demonstrates our model performance.
Fig. 4 Original image and the predicted mask image
Fig. 5 Example of edema region in gliomas
Table 2 Dice coefficient of
three sub-regions of gliomas
tumor
Sub-regions Dice coefficient
Full tumor 0.68
Tumor core 0.79
Enhancing tumor 0.75
Brain Tumor Segmentation Using U-Net 159
Fig. 6 Comparison between training and validation of accuracy, loss, Dice coefficient, and the IOU
5 Conclusion
In this study, we present a more advanced version of the U-Net design for segmenting
brain tumors. Based on studies with well-known benchmarking datasets (BRATS
2020), we have proven that, when compared to manually delimited ground truth,
they can deliver efficient and reliable segmentation. Our U-Net deep convolutional
networks, which are entirely based on U-Net, can also obtain comparable results for
the entire tumor region and superior results for the core tumor region.
References
1. Satyanarayana KV, Rao NT, Bhattacharyya D, Hu Y (2022) Identifying the presence of
bacteria on digital images by using asymmetric distribution with k-means clustering algorithm.
Multidimension Syst Signal Process 33(2):301–326. https://doi.org/10.1007/s11045-021-008
00-0
2. Asish Vardhan K, Thirupathi Rao N, Naga Mallik Raj S, Sudeepthi G, Divya, Bhattacharyya
D, Kim T (2019) Health advisory system using IoT technology. Int J Recent Technol Eng
7(6):183–187. Retrieved from www.scopus.com
3. Eali SNJ, Bhattacharyya D, Nakka TR, Hong S (2022) A novel approach in bio-medical image
segmentation for analyzing brain cancer images with U-NET semantic segmentation and TPLD
models using SVM. Traitement Du Signal 39(2):419–430. https://doi.org/10.18280/ts.390203
4. Sarasaen C et al (2021) Fine-tuning deep learning model parameters for improved super-
resolution of dynamic MRI with prior-knowledge. Artif Intell Med 121:102196
5. Pallud J et al (2012) Quantitative morphological magnetic resonance imaging follow-up of
low-grade glioma: a plea for systematic measurement of growth rates. Neurosurgery 71:729–
740
6. Eali SNJ, Rao NT, Swathi K, Satyanarayana KV, Bhattacharyya D, Kim T (2018) Simulated
studies on the performance of intelligent transportation system using vehicular networks. Int J
Grid Distrib Comput 11(4):27–36. https://doi.org/10.14257/ijgdc.2018.11.4.03
7. Rajpurkar P et al (2018) Deep learning for chest radiograph diagnosis: a retrospective
comparison of the CheXNeXt algorithm to practicing radiologists. PLoS Med 15:e1002686
8. Ge C et al (2018) Deep learning and multi-sensor fusion for glioma classification using multi-
stream 2D convolutional networks. In: 2018 40th Annual international conference of the IEEE
engineering in medicine and biology society (EMBC), pp 5894–5897
160 P. Jyothsna et al.
9. Joshua ESN, Bhattacharyya D, Chakkravarthy M, Kim H (2021) Lung cancer classification
using squeeze and excitation convolutional neural networks with grad cam++ class activation
function. Traitement Du Signal 38(4):1103–1112. https://doi.org/10.18280/ts.380421
10. Pérez-García F et al (2021) Torchio: a python library for efficient loading, preprocessing,
augmentation and patch-based sampling of medical images in deep learning. Comput Methods
Programs Biomed 2021:106236
11. Chatterjee S, Nizamani FA, Nürnberger A et al (2022) Classification of brain tumours in MR
images using deep spatiospatial models. Sci Rep 12:1505. https://doi.org/10.1038/s41598-022-
05572-6
An Empirical Study of CNN-Deep
Learning Models for Detection
of Covid-19 Using Chest X-Ray Images
Mohd. Abdul Muqeet, Quazi Mateenuddin Hameeduddin,
B. Mohammed Ismail, Ali Baig Mohammad, Shaik Qadeer,
and M. Muzammil Parvez
Abstract The Covid-19 spun into a pandemic and has affected routine lives and
global health. It is crucial to identify the infectious Covid-19 subjects as early as
possible to avert its spread. The CXR images processed with deep learning (DL)
processes have newly become an earnest method for early Covid-19 detection along
with the regular RT-PCR test. This paper examines the deep learning (DL) models to
detect Covid-19 from CXR images for early analysis of Covid-19. We conducted an
empirical study to assess the efficacy of the proposed convolutional neural network
DL model (CNN-DLM), pre-trained with some eminent networks such as MobileNet,
InceptionNet-V3, ResNet50, Xception, and DenseNet121 for initial detection of
Covid-19 for an openly accessible dataset. We also exhibited the accuracy and loss
value curves for the selected number of epochs for all these models. The results
indicate that with the proposed CNN model pre-trained with the DenseNet121 greater
Mohd. A. Muqeet (B)
Electrical Engineering Department, Muffakham Jah College of Engineering and Technology,
Hyderabad, India
e-mail: ab.muqeet2013@gmail.com
Q. M. Hameeduddin
Faculty of Electronics and Communication Engineering, Indian Naval Academy, Ezhimala,
Kerala, India
B. Mohammed Ismail
Department of Artificial Intelligence & Machine Learning, P.A. College of Engineering,
Mangalore, Karnataka, India
A. B. Mohammad
School of Electronics and Communication Engineering, REVA University, Bengaluru, India
S. Qadeer
Electrical Engineering Department, Muffakham Jah College of Engineering and Technology,
Hyderabad, India
M. Muzammil Parvez
Electronics and Communication Engineering Department, KLEF, Deemed to Be University,
Vaddeswaram, A.P, India
e-mail: parvez@kluniversity.in
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_17
161
162 M. A. Muqeet et al.
results were achieved compared to other pre-trained CNN-DLMs applied in a transfer
learning approach.
Keywords Covid-19 ·Chest X-Ray (CXR) Images ·Convolutional neural
network (CNN) ·Deep learning models (DLM)
1 Introduction
Due to the amount of Covid-19 spread which turned into a severe medical issue
considered a vastly transmissible disease [1]. The RT-PCR trials are standard inves-
tigations for Covid-19, which examine the existence of antibodies of a specified
disease [2]. But due to its manual examination process and usage of a testing kit, it
tends to be a slow process with an accuracy of not more than 64% [3]. Furthermore,
in RT-PCR trials, other significant techniques that facilitate the detection of Covid-19
include chest radiography [4, 5] which assists in the timely detection for immediate
medical intervention. The dissimilarity of the CXR image of a normal and Covid-19
infected subject is illustrated in Fig. 1.
Numerous researchers have focused their studies to apply artificial and deep
learning techniques for medical image analysis [6, 7]. To discover Covid-19 in
CXR images, COVIDX-Net [8] model was suggested using CNN. DarkCovidNet
[9] was developed based on the CXR images for automatic Covid-19 diagnosis and
realized a binary classification accuracy of 98.08%. Prevailing CNN models were
applied for Covid-19 patient classification and attained an accuracy of 98.75% [10].
Using the CXR images, Narin et al. [11] used the ResNet50 CNN-DLM to achieve
98% Covid-19 accuracy for 100 images (50 normal and 50 Covid-19). Muqeet et al.
[12] demonstrated good performance using features retrieved from CXR images
Fig. 1 Variation between the CXR image of a normal patient and Covid-19 patient
An Empirical Study of CNN-Deep Learning Models for Detection … 163
after applying a few prominent DL models. COVIDiagnosis-Net using SqueezeNet
[13] was proposed along with a Bayesian optimizer, and an accuracy of 98.30%
was reported. Farooq and Hafeez [14] applied ResNet-based CNN-DLM termed
as COVIDResNet for Covid-19 classification and reported an accuracy of 96.23%.
Similarly, recent investigations [15, 16] also exhibit the advantage of DL methods for
the detection of the Covid-19. Section 2 explores the details of the applied database
and also discusses the steps for model formulation for Covid-19 detection process.
Section 3 describes the empirical studies conducted on different CNN-DLMs and
suggests the best suitable CNN-DLM for the detection task. Lastly, in Sect. 4,the
paper is concluded.
2 Materials and Methods
The next section outlines the dataset selection and also provides the details of model
formulation and CNN-DLM structure applied for the detection of Covid-19 from
CXR images.
2.1 Dataset
There are some data sources accessible through sources like Github and Kaggle
with CXR images. The dataset is considered from the Github repository [17] which
includes CXR images from both normal and Covid-19 patients.
2.2 Model Formulation
With the help of transfer learning, this CNN network also demonstrates a strong
ability to generalize the images external to the ImageNet dataset [18, 19]using the
Keras core library. There are various pre-trained CNN-DLMs available for image
classification. Among these, we have considered MobileNet [20], Inception-V3 [21],
ResNet50 [22], Xception [23], and DensNet121 [24] in a transfer learning approach
for the detection of Covid-19 patient. The CNN-DLM with a layered network is much
useful to capture the characteristics of the images. The following steps illustrate the
building procedure of our proposed CNN-DLM. The proposed model is applied in a
transfer learning approach to the selected pre-trained models.
(1) Initially, the CXR images of the normal and Covid-19 subjects are applied as
input to the CNN-DLM.
(2) Convolution filters and feature maps are applied to each of the individual images,
which generate a convolution layer.
164 M. A. Muqeet et al.
(3) Next, MaxPooling and ReLU functions are applied at each end of the convo-
lution layer to accomplish the nonlinear transformation of inputs present in the
model. The ReLU activation layer offers the CNN model further acceleration
to perform an additional complex task.
(4) Following resultant images are forwarded to the Pooling layer to offer spatial
invariance to the CNN-DLM and to produce pooled feature map.
(5) Dropout layer with 0.5 dropout rate is applied to reduce the overfitting of the
model.
(6) The Steps 2–5 are applied 2 more times.
(7) The final pooled feature map attained in earlier step is flattened.
(8) Finally, a Dense layer with Softmax activation function is applied to execute
the classification of a binary output using the binary cross-entropy function.
3 Experimental Results
The outcome of the empirical study on various CNN-DLMs for the selected dataset
is discussed here. We consider the dataset from [17] with 1800 images in total to
evaluate the suggested technique. For the training process, 80% and, for evaluation,
20% of the CXR images in this dataset are used. The CXR image is classified as
Covid-19 infected or normal using a DL-based classification algorithm based on this
data. The work is carried out with the open-source Keras [19] framework and the
TensorFlow [20] backend. The proposed work is implemented using Google Colab.
3.1 Evaluation Matrices
A confusion matrix [16] is an outline of detection results for a classification process
that discusses parameters such as recall, precision, F1-Score, and accuracy [25]. The
performances of the developed model are evaluated based on some estimated values.
True positives (tp) are those where actual and predicted results are positive. False
negatives ( fn) are those where actual results are positives but predicted results are
negative. True negatives (tn) are those where actual and predicted results are nega-
tives. False positives ( fp) are those where actual results are negatives but predicted
results are positives. A test’s recall is specified as:
Recall =tp
tp + fn
(1)
The test’s precision is specified as:
Precision =tp
tp + fp
(2)
An Empirical Study of CNN-Deep Learning Models for Detection … 165
The correctness of the result is also indicated by the F1-Score specified as:
F1-Score = 2 × [Recall × Precision]
[Recall + Precision] (3)
The accuracy of the experiment in terms of confusion matrix parameters can be
specified in (Eq. 4) as follows:
Accuracy =tp + tn
tp + fn + fp + tn
(4)
We also applied some hyper-parameters value settings. For data augmentation,
we considered the rotation range to be 15. The Adam optimizer is selected with a
batch size of 8. The epoch is selected as 10 uniformly for all the comparative models.
3.2 Comparison of Various CNN-DL Models
We presented the confusion matrix results of five pre-trained CNN-DLMs with
proposed model on the selected dataset. The training performance in terms of training
loss, validation loss, and validation accuracy is reported. Table 1 shows the recall,
precision, F1-Score, and accuracy values for the applied pre-trained CNN-DLMs for
detection of Covid-19 CXR images.
It is noticed that the proposed model pre-trained with DenseNet121 model attained
the top outcomes with a precision of 100%, recall of 98%, F1-Score of 99%, and
accuracy of 99.45%. Figures 2, 3, 4, 5 and 6 illustrate the training loss and accuracy
plots for the selected CNN-DL models throughout 10 epochs. Figure 7 illustrates the
results of proposed model pre-trained with the DenseNet121. The indicator true value
= 0 indicates that the detection = 0 for Covid-19 correct result, whereas true value
= 1 indicates that the detection = 1 for normal patients with correct outcome. The
special structure of this DenseNet121 model enhances data flow across the network
and improves parameter efficacy.
Table 1 Different CNN-DLMs and parameter specifications
CNN-DL models Precision Recall F1-Score Accuracy
MobileNet 100 97.45 98.10 98.89
InceptionNet-V3 98.63 98.30 98.08 98.33
ResNet50 97.87 96.43 97.27 97.77
Xception 100 96.11 98.19 98.61
DenseNet121 100 98.20 99.04 99.45
166 M. A. Muqeet et al.
Fig. 2 Plots of the MobileNet model
Fig. 3 Plots of the InceptionNet-V3 model
Fig. 4 Plots of the ResNet50 model
3.3 Comparison of Various CNN-DL Models
Here, we compared the most excellent CNN-DLM results with newly developed DL
methods for Covid-19 detection using CXR images as tabulated in Table 2. It is noted
An Empirical Study of CNN-Deep Learning Models for Detection … 167
Fig. 5 Plots of the Xception model
Fig. 6 Plots of the DenseNet121 model
that the proposed work attained better results compared with other existing methods.
Compared to notable work in [8] and [11], we considered a relatively larger number
of CXR images to test CNN-DLMs. The studies mentioned in [9, 13, 14], and [15]
applied a relatively larger datasets to test their models, but these datasets suffered
from class imbalance issues and compromised with a reduced number of Covid-19
CXR images, whereas, in our study, the dataset has appropriate class distribution for
Covid-19 and normal CXR images. It is also noted that the proposed work which is
the proposed CNN-DLM with the pre-trained model selected as DenseNet121 attains
a 100% Covid-19 precision value.
4 Conclusion
The work aims to develop the CNN-DLM-based detection of Covid-19 from CXRIs
images. A larger dataset of CXRIs is considered. A CNN model is proposed which is
pre-trained with prominent CNN-DLMs. The results indicated that the proposed
CNN-DLM pre-trained with DenseNet121 surpassed other CNN-DLMs with an
168 M. A. Muqeet et al.
Fig. 7 Detection Results for a few CXR Images (DenseNet121)
Table 2 Covid-10
performance comparison with
notable work
CNN-DL methods Covid-19 class precision
Hemdan et al. [8]100
Narin et al. [11]96
Ozturk et al. [9]90.65
Ucar and Korkmaz [13]100
Farooq and Hafeez [14]100
Wang and Wong [15]87.10
Proposed Work 100
An Empirical Study of CNN-Deep Learning Models for Detection … 169
accuracy of 99.45% for detection of Covid-19. The other confusion matrix parameters
are also better compared with other pre-trained CNN-DLMs. Thus, this empirical
study suggested the possible application of DL techniques for early detection of
Covid-19.
References
1. World Health Organization (2020) Coronavirus disease 2019 (COVID-19) Situation Report–
196 [cited 3 Aug 2020]
2. Rahmani AM et al (2022) Automatic COVID-19 detection mechanisms and approaches from
medical images: a systematic review. Multimed Tools Appl. https://doi.org/10.1007/s11042-
022-12952-7
3. Agrawal T, Choudhary P (2021) FocusCovid: automated COVID-19 detection using deep
learning with chest X-ray images. Evol Syst. https://doi.org/10.1007/s12530-021-09385-2
4. Durrani M, Haq I u, Kalsoom U, Yousaf A (2020) Chest X-rays findings in COVID 19 patients
at a University Teaching Hospital—A descriptive study. Pakistan J Med Sci 36 (COVID19-
S4):S22
5. Sakib S et al (2022) Detection of COVID-19 disease from chest X-ray images: a deep transfer
learning framework. MedRxiv. https://doi.org/10.1101/2020.11.08.20227819
6. Nyemeesha V, Ismail BM (2021) Method to enhance classification of skin cancer using back
propagated artificial neural network applied intelligence and informatics. AII 2021. Commun
Comput Inf Sci 1435
7. Ismail M et al (2019) An effective heart disease prediction method using artificial neural
network. Int J Innov Technol Exploring Eng 8(8):1529–1532
8. Hemdan EED, Shouman MA, Karar ME (2020) COVIDX-Net: a framework of deep learning
classifiers to diagnose COVID-19 in X-Ray Images, arXiv preprint arXiv: 2003.11055
9. Ozturk T, Talo M, Yildirim EA, Baloglu UB, Yildirim O, Rajendra AU (2020) Automated
detection of COVID-19 cases using deep neural networks with X-ray images. Comput Biol
Med 121:103792
10. Apostolopoulos ID, Mpesiana TA (2020) Covid-19: automatic detection from X-ray images
utilizing transfer learning with convolutional neural networks. Phys Eng Sci Med 43(2):635–
640
11. Narin A, Kaya C, Pamuk Z (2021) Automatic detection of coronavirus disease (COVID-19)
using X-ray images and deep convolutional neural networks. Pattern Anal Appl 24(3):1207–
1220
12. Muqeet MA, Quadri MU, Sasidhar K, Krishna PS (2022) Deep learning-based prediction of
nCOVID-19 disease using chest X-ray images (CXRIs). In: Chaurasia MA, Mozar S (eds)
Contactless healthcare facilitation and commodity delivery management during COVID 19
pandemic. Advanced technologies and societal change. Springer, Singapore. https://doi.org/
10.1007/978-981-16-5411-4_3
13. Ucar F, Korkmaz D (2020) COVIDiagnosis-Net: deep Bayes-SqueezeNet based diagnosis of
the coronavirus disease 2019 (COVID-19) from X-ray images. Med Hypotheses 140:109761
14. Farooq M, Hafeez A (2003) COVID-ResNet: a deep learning framework for screening of
COVID19 from radiographs, 2020, arXiv preprint arXiv: 2003.14395
15. Wang A, Wong, COVID-Net: a tailored deep convolutional neural network design for detection
of COVID-19 cases from chest X-Ray images, 2020, arXiv preprint arXiv:2003.09871
16. Sethy PK, Behera SK, Detection of coronavirus disease (COVID-19) based on deep features,
2020. https://doi.org/10.20944/preprints202003.0300.v1
17. Cohen JP, Morrison P, Dao L (2020) COVID-19 image data collection. https://github.com/iee
e8023/Covid-chestxray-dataset
170 M. A. Muqeet et al.
18. Shi F et al (2020) Review of artificial intelligence techniques in imaging data acquisition,
segmentation and diagnosis for COVID-19. https://arxiv.org/abs/2004.0273
19. Krizhevsky A, Sutskever I, Hinton GE (2012) ImageNet classification with deep convolutional
neural networks. In: Advances in neural information processing systems, pp 1097–1105
20. Sandler M, Howard A, Zhu M, Zhmoginov A, Chen L-C (2018) MobileNetV2: Inverted resid-
uals and linear bottlenecks. In: Proceedings of the IEEE conference on computer vision and
pattern recognition, 2018, pp 4510–4520
21. Szegedy C, Vanhoucke V, Ioffe S, Shlens J, Wojna Z (2016) Rethinking the inception archi-
tecture for computer vision. In: Proceedings of the IEEE conference on computer vision and
pattern recognition, 2016, pp 2818–2826
22. He K, Zhang X, Ren S, Sun J (2016) Deep residual learning for image recognition, in:
Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp
770–778
23. Chollet F (2017) Xception: deep learning with depth wise separable convolutions. In
Proceedings of the IEEE conference on computer vision and pattern recognition 2017, pp
1800–1807
24. Huang G, Liu Z, Van Der Maaten L, Weinberger KQ (2017) Densely connected convolutional
networks. In: Proceedings of the IEEE conference on computer vision and pattern recognition,
Honolulu, 21–26 July 2017, pp 4700–4708
25. Muqeet MA, Holambe RS (2018) A collaborative representation face classification on separable
adaptive directional wavelet transform based completed local binary pattern features. Eng Sci
Technol, Int J 21(4)
Detection of Eye Blink Using SVM
Classifier
Varaha Sai Adireddi, Charan Naga Santhu Jagadeesh Boddeda,
Devi Shanthisree Kumpatla, Chris Daniel Mantri, B. Dinesh Reddy ,
G. Geetha, N. Thirupathi Rao , and Debnath Bhattacharyya
Abstract The eyes are the most important feature of our bodies because they allow
us to see and explore the world. Nowadays, technology is continually evolving,
paving the way for greater development and increased use of gadgets by everyone.
When people stare at digital screens for long periods of time, they develop eye strain
and visual issues, which is known as computer vision syndrome (CVS). The best way
to avoid visual problems caused by digital screens is to take appropriate preventive
measures such as getting regular eye care. To protect users from eye disorders, we
created a model that uses the Viola–Jones method and the SVM classifier to estimate
the user’s eye blinking ratio. As a result, the proposed approach calculates locations
of significance and another scalar parameter is derived—ratio of the eyes (EAR)—
that characterizes each frame’s eye opening. Finally, in a limited temporal window,
eye blinks are recognized as a pattern of EAR values using an SVM classifier. The
user can be notified about his gadget usage based on the results of the eye blink ratio
and gradually diminish his addiction to digital screens that affect his eyes.
Keywords Computer vision syndrome ·Eye blink ·Eye aspect ratio ·Viola–Jones
algorithm ·Facial landmarks ·Support vector machine
V. S. Adireddi (B) · C. N. S. J. Boddeda · D. S. Kumpatla · C. D. Mantri · B. D. Reddy ·
N. T. Rao
Department of Computer Science and Engineering, Vignan’s Institute of Information Technology
(A), Visakhapatnam, Andhra Pradesh, India
e-mail: varahasaiadireddi@gmail.com
G. Geetha
Department of Information Technology, VR Siddhartha Engineering College, Kanuru,
Vijayawada, Andhra Pradesh, India
D. Bhattacharyya
Department of Computer Science and Engineering, Koneru Lakshmaiah Education,
Vaddeswaram, Guntur, Andhra Pradesh 522502, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_18
171
172 V. S. Adireddi et al.
1 Introduction
Computers have become an inextricable aspect of our lives. The contemporary age is
increasingly spending hours together in front of visual technologies such as a personal
digital assistant, computer monitor, and television. One of the adverse consequences
of this thinking is computer vision syndrome (CVS), often known as digital eye strain.
CVS is a collection of visual and eye disorders caused by long periods of sitting in
front of computers. According to a recent poll, India’s computer population is above
20 million, with 80% of them suffering from CVS. CVS symptoms affect 25–93%
of computer users. Researchers have written about CVS in medical and technical
journals. The task’s visual demands frequently exceed an individual’s visual capacity
to do them comfortably, resulting in CVS symptoms. Those who spend two or more
hours every day in front of a computer or on a digital screen gadget are most likely
to develop CVS. CVS lowers the blink rate (number of blinks per unit of time) and
causes the majority of the difficulties. As a result, the best CVS approach might
be to concentrate on preserving the user’s nominal blink rate on an average time
scale. This is the driving force behind this project. A blink has been stimulated by
researchers. The blink detection/rate could also be used to reduce accidents in driver
awareness/alertness systems. Apart from that, the blink is used in computer eye-based
contact and communication. Non-intrusive methods, s uch as using a camera to detect
eye blinks, have been the subject of recent research. They have challenges with blink
detection accuracy, eye shape variation, and user mobility, despite promising results.
We provide a new approach for identifying eye blinks based on EAR ratios predicted
using an SVM classifier, which accurately gives the eye blink ratio so that the user
can quickly access his eye condition and reduce the use of digital screen devices in
this study.
2 Related Works
As we mentioned earlier, they are several papers produced on eye blink detection
concepts. We had reviewed several technologies mentioned in the literature. Concept
of feature selection is used to extract the facial landmarks from the images using a
pre-trained model which extracts the face from an image (Haar) Cascade function.
The detection framework is always looking for the image pixel sums within rectan-
gular rectangles. As a result, they resemble Haar basis functions, which have been
employed in image-based object detection in the past. On the other hand, they are
more intricate due to the fact that they all rely on more than one rectangular area.
Subtract the total of the pixels inside clear rectangles from the sum of the pixels inside
shaded rectangles to get the value of each feature. When compared to alternatives
like steerable filters, rectangular features like this are rudimentary. They respond to
horizontal and vertical elements, but their response is significantly coarser [1–4].
Detection of Eye Blink Using SVM Classifier 173
The use of intensity vertical projection to detect eye blinks, which estimates the
total intensity of object pixels in each row. According to them, the brow and iris
parts should be darker than the skin. As a result, two IVP local minima will represent
their centers. The center of the skin area between the eyebrow and the iris is the
maximum distance between them. Because the skin area expands when the eyes are
closed, the minimum distance is now the middle of the eyelids when the eye is closed
[5–9]. Another theory proposed a 3-stage method where it contains three stages, the
first of which introduces an integral image that serves as an intermediate image.
Second, the AdaBoost method is used to extract key characteristics. All complex
features are integrated in the final stage to get a clear face detection. Second, they
employed eye pair detection, which detects the eyes using Golden ratios. It produces
templates from the frame in this case. After the eye tracking template matching is
accomplished, track the eyes and obtain great accuracy [10].
The eye blinking can be detected via motion analysis. A quick eye tracking process
preserves perfect information of the eye’s appearance after first localization. Motion
analysis provides incredibly precise facts regarding the eyes’ locations when blink
detection is used. As a result, updating the area of interest (ROI) centered around the
eye requires only a basic tracking algorithm. The normalized correlation coefficient is
used by the system. Based on the user’s eye blink rate, the eye protection programmed
assesses whether or not the user should rest their eyes. It accurately recognizes the
eyeballs from a variety of viewing angles and lighting situations.
3 Materials and Methods
3.1 Face Detection
In the first stage, the Viola–Jones face detector is used to recognize the face. It is
a real-time object detection approach with a high detection rate. There are three
essential steps to it. First, an intermediate image in the form of an integral image
is introduced to speed up feature extraction by employing pixel sums rather than
rectangular features, which are regarded to be slow. Second, the AdaBoost technique
is utilized to extract key characteristics from a large amount of data, yielding a highly
accuracy (Fig. 1).
3.2 Eye Pair Detection
Following the detection of the face, the eyes are detected in the second step. As we
all know, the eye pair is located in the upper area of the face (EAR). The iBUG
300-W dataset, which comprises 68 face landmarks, was used to train the dlib facial
174 V. S. Adireddi et al.
Fig. 1 Facial landmarks
Fig. 2 Eye facial landmark
points
landmark predictor. We can detect the pair of eyeballs indicated by 6 points using
the landmarks points (Fig. 2).
3.3 Classification Using SVM Model
Eyeblink8 is the name of the dataset, which consists of eight films and four individuals
which is used for training the SVM model. Many non-blink motions, natural face
movements, and facial mimics are presented. There are over 82,600 frames (640,480)
in the collection, with 353 blinks. The dataset distinguishes between three states:
open, half, and closed. When the blink begins, individual frames are allocated half
tags until the blink is entirely closed. The term “fully closed eye” refers to when the
eyelid covers 90–100% of the eye. Close is used to tag entirely closed eyes, whereas
half is used to tag opened eyes till they are fully open. This method can also be used
to mark eye blinks that are not completely closed. The tag Left/Right is added to the
eye state if only one eye is visible.
As a result, we present a classifier that uses a frame’s broader temporal window
as an input. In 30 fps recordings, we discovered that three frames can have a signif-
icant impact on blink recognition for a frame where an eye is the most closed
Detection of Eye Blink Using SVM Classifier 175
during blinking. Concatenating the EARs of its three adjoining frames yields a
7-dimensional feature for each frame.
The data frame given input to SVM is. Based on the data frame, the SVM classifier
finds the y value as yk = 1 if at least one of the frames in the 7-dimensional window
is totally closed, and yk = 0 if all of the frames in the same 7-dimensional window
are half or open.
Manually labeled sequences y were used to train the linear SVM classifier. For
each frame, save the three frames at the beginning and end of a video sequence, and
a 7-dimensional feature x1, …, x7 is generated (using the EAR measure) and classed
by SVM.
Given the enormous amount of 0s in y, the idea was to balance the dataset by
picking a random sample of 0s equal to the number of 1s for each movie. The end
result was a dataset with 5900 rows that was randomly split into two sets: training
(80% of the observations) and test (20% of the observations). The two sets have an
equal distribution of 0s and 1s: The training set has 2348 1s across 4720 units, while
the test set has 602 1s over 1180 units.
To avoid differing EAR scales when comparing multiple films, a normalization
was performed first for each video (before s ampling the 0’s and 1’s) and then for the
training set. It should be mentioned that SVM detects the presence of closed eyes in
the 7-dimensional window with 95% accuracy. As a result, when a blink happens,
the SVM precision is made by a sequence of seven 1’s between 0’s, assuming that
the classifier is never erroneous and that the eyes are closed for only one frame. This
sequence must be reduced to a single blink. It is vital to remember that both false
positives and false negatives can cause SVM predictions to be incorrect (Fig. 3).
Following considerable research, we discovered that the optimal rule was to
produce only sequences of turning t he only 1’s sequence into blinks by repeating the
1’s and 0’s. The sequences of consecutive 1 s and 0 s are made as follows:
Fig. 3 Finding the EAR values using SVM classifier
176 V. S. Adireddi et al.
Table 1 Comparing results between SVM and EAR thresholding
SVM OpenCV
Precision (%) Recall (%) Precision (%) Recall (%)
Video 1 43 81 712
Video 2 100 23 0 0
Video 3 58 89 733
0’s: single (…, 1, 1, 0, 1, 1), double (…, 1, 1, 0, 0, 0, 1, 1), and triple (…, 1,
1, 0, 0, 0, 0, 1, 1) (…, 1, 1, 0, 0, 0, 1, 1, …) 0’s in between consecutive numbers.
Misclassified 1’s were identified and reclassified as 1.
1’s: single (…, 0, 0, 1, 0, 0, …) and double (…, 0, 0, 1, 1, 0, 0, …) 1’s between
series of consecutives 0’s were identified as misclassified and changed to 0’s.
A blink is detected after this transition when a sequence of 1’s is found.
4 Results and Discussions
The performance was assessed not only in terms of the number of correctly catego-
rized blinks (true positives), but also in terms of the significant occurrences in which
a video frame was misclassified as a blink (false positives) or when a blink was
missed (missed blinks) (false negatives). Two metrics have been generated based on
these data.
After being confirmed on multiple videos from the iBUG 300-W dataset, the SVM
blink detector was compared against the OpenCV blink detector. The goal of this
step was to see how well the two systems could detect blinks despite differences in
stance, expression, lighting, backdrop, occlusion, and image quality. The EAR SVM
experiment is conducted across multiple datasets. The Eyeblink8 dataset is used to
train the SVM classifier, which is then evaluated on the Talking dataset (Table 1).
We believe these results are relevant and SVM classifier is far better than OpenCV
model (Fig. 4).
We find that EAR thresholding is behind both EAR SVM classifiers in this tough
database. The thresholding fails when a subject grin looks to the side or closes his
or her eyes for longer than a blink duration. The SVM detector greatly outperforms
the EAR thresholding value.
5 Conclusion
A real-time t echnique for identifying eye blinks was presented. We proved quanti-
tatively that correlation coefficients’ facial feature classifiers are accurate enough to
estimate eye openness consistently. Under all tough conditions, the OpenCV blink
Detection of Eye Blink Using SVM Classifier 177
Fig. 4 Comparative analysis of SVM and OpenCV models
detector performs poorly. When dealing with facial emotions and position fluctua-
tions, the OpenCV detector has a number of issues, including misclassifying blinks
under random settings. On two standard datasets, we developed a model that employs
a state-of-the-art robust landmark detector, followed by a basic SVM-based eye
blink detection. Because the additional processing expenses for eye blink detection
are modest compared to the real-time landmark detectors, the algorithm works in
real time. The suggested SVM approach outperforms EAR thresholding by using a
temporal window of the eye aspect ratio (EAR). In the event that a lengthier sequence
is not available, the thresholding can be used as a single image classifier to detect
the eye state. In the future, we can deploy this classifier with the front camera of the
Android device to calculate the activity time of the user and notify him easily.
References
1. Sandberg D et al (2011) The characteristics of sleepiness during real driving at night—a study
of driving performance physiology and subjective experience. Sleep 34(10):1317
2. Lin CT, Chang CJ, Lin BS, Hung SH, Chao CF, Wang IJ (2010) A real-time wireless
brain—computer interface system for drowsiness detection. IEEE Trans Biomed Circuits Syst
4(4):214–222
178 V. S. Adireddi et al.
3. Lin CT, Chen YC, Huang TY, Chiu TT, Ko LW, Liang SF et al (2008) Development of wireless
brain computer interface with embedded multitask scheduling and its application on real-time
driver’s drowsiness detection and warning. IEEE Trans Biomed Eng 55(5):1582–1591
4. Bhattacharyya D, Doppala BP, Thirupathi Rao N (2020) Prediction and forecasting of persistent
kidney problems using machine learning algorithms. Int J Current Res Rev 12(20):134–139.
https://doi.org/10.31782/IJCRR.2020.122031
5. Mandhala VN, Bhattacharyya D, Vamsi B, Thirupathi Rao N (2020) Object detection using
machine learning for visually impaired people. Int J Current Res Rev 12(20):157–167. https://
doi.org/10.31782/IJCRR.2020.122032
6. Picot A, Charbonnier S, Caplier A (2008) On-line automatic detection of driver drowsiness
using a single electroencephalographic channel. In: Proceedings of 30th annual international
conference of the engineering in medicine and biology society 2008. EMBS IEEE
7. You C-W et al (2012) CarSafe: a driver safety app that detects dangerous driving behavior using
dual-cameras on smartphones. In: Proceedings of the 2012 ACM conference on ubiquitous
computing
8. Kurian D, Johnson Joseph PL, Radhakrishnan K, Balakrishnan A (2014) Drowsiness detec-
tion using photoplethysmography signal. In: Fourth International conference on advances in
computing and communications (ICACC), pp 73–76
9. Sahayadhas A, Sundaraj K, Murugappan M (2013) Drowsiness detection during different times
of day using multiple features. Australas Phys Eng Sci Med 36(2):243–250
10. Pal NR et al. (2008) EEG-based subject-and session-independent drowsiness detection: an
unsupervised approach. EURASIP J Adv Signal Process
A Novel Approach for Health Analysis
Using Machine Learning Approaches
Debdatta Bhattacharya , N. Thirupathi Rao , K. Asish Vardhan,
and Eali Stephen Neal Joshua
Abstract Data mining and big data are today the world’s leading technology. These
techniques deal with diabetes in the banking sector, health services, cyber-security,
voting, insurance, the real state, etc. Diabetes is a constant disease before digestion,
and wherever personality and total amount in the body of blood glucose is experi-
enced, the formation of estrogens is also unsatisfactory, otherwise the carcass phones
do not react properly to estrogens. The balance in high blood sugar diabetes is noto-
rious for extensive stretch injuries, twitching, difficulty’s evolutionary structure of
kidneys, heart, vein, nerves and eyes in particular. That is, the main purpose is to
analyze consumption, plan a predictable outcome, using the technique of machine
learning and position the classifying model with a medical outcome to the adjacent
effect. The system mainly selects the features that make miserable diabetes mellitus
in the early detection of extrapolative studies. Different results algorithms display
the random forest as well as the decision tree algorithm with the greatest distin-
guishability of 97.20 and 97.30%. Discreetly, diabetics perform best inspection of
information. Information. Naive Bayesian has an optimal outcome of precision of
85.43%. Similarly, the study provides a summary of the model highlights selected
to develop the data collection precisely.
Keywords SVM ·Diabetes ·Naive Bayesian ·Random forest ·Data mining ·Big
data ·Machine learning ·Deep learning
D. Bhattacharya (B)
Department of Computer Science and Engineering, Koneru Laksmaiah Education Foundation,
Vaddeswaram, Guntur, Andhra Pradesh 522302, India
e-mail: debdatta122001@gmail.com
N. T. Rao · E. S. Neal Joshua
Department of Computer Science and Engineering, Vignan’s Institute of Information Technology,
Visakhapatnam, Andhra Pradesh 530016, India
K. Asish Vardhan
Department of Computer Science and Engineering, Bullayya College of Engineering for Women,
Visakhapatnam, Andhra Pradesh, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_19
179
180 D. Bhattacharya et al.
1 Introduction
Mainly, The World Health Association of annual statement indicates an amount of
diabetes people encounter is 434 million per year (note down starting from first
year to last year this is very important imminent for 434 million). Constantly, here
some important adding up for quantity of people encountering diabetes in a variety of
restoration center of attention. The World Health Organization (WHO) statements on
“Diabetes Care 2018” for medical consideration in Diabetes by American Diabetes
Association and Standards [1, 2], an assessment of relation various aspects as well as
their reimbursement. Figure 1 depicts various people (sexual orientation and reim-
bursement) developed someplace in the collection of 30–80 years, the transient of
levels of hypertension.
The constant of diabetes mellitus [3] is constant difficulty anywhere its reason
on description of the circulatory structure by high sugar levels. It is the reason of
the pancreatic beta cells due to the erroneous functioning. This affects different
pieces of the body which incorporates pancreas problem, kidney disillusionments,
pancreatic issues, hypertension, foot issues, nerve hurt, threat of heart diseases, eye
issues, ketoacidosis, glaucoma, visual agitating impacts and cascades, etc., in the rear
justification such as standard of living of a man due to different purposes, significant
cholesterol (Hyperlipidaemia), so the deficiency of the movement, s moking, strength,
nutrition propensities, hypertension (Hyperglycaemia) and so on which is extremely
adding the risk of treating fundamental levels of diabetes. These impacts on a large
scale of age, as well as young people to mature and developed individually.
Pancreas [4] is a limb set in the waist region. It has two necessary restrictions; the
first one is endocrine bound and the second one is exocrine bound. This endocrine
helps the incorporation as well as the exocrine part of the pancreas keeps up the flow
structure in the sugar level. Recognition of the pancreas is with impacts from different
Fig. 1 Finding the diabetes levels and their ranges which consolidates the different pieces of the
body is affected by diabetes
A Novel Approach for Health Analysis Using Machine … 181
pieces of the body and different insufficiency [4]. In the body, the confirmation of
sugar levels is like way expects in diabetes of fundamental movement.
The thought of failure is where the retinopathy retina, optic nerve and retina
position of the meeting be injured. An after effect of terminate issues for the night-
time illustration debilitation, increasing the retina area; the awareness of reducing
the contact might be occurred. A pair of tests near the initial occasions [5] should
handle the diabetic individual eye visualization during pharmaceutical. Consolidates
the treatment for image unevenness testing, optic comprehensibility tomography
(OCT), alternate growth and to none attempt. The treatment joins diverse medicines,
corticosteroid, middle/traverse piece macular laser restorative methodology and anti-
VEGF implantation.
2 Related Works
The diabetes does not have the infectious illness be prompting extensive pull incon-
veniences and real health issues. The proof comes by World Health Organization
expresses about the diabetes with complexities of outcome occurred by person
mentally, economically and economically in excess of their positions. Our anal-
ysis said regarding 2.1 million passing’s because of their unrestrained health period
show toward passing away. As regards 3.5 million passing’s [6] happened because of
the danger mechanism of diabetes similar to congestive heart failure with dissimilar
maladies.
The diabetes nothing but a sickness which is the reason because the comprehensive
study of sugar levels is fixation into the blood. In literature study, examined dissim-
ilar analysis, alternative expressively encouraging classification is recommended
utilizing the AdaBoost computation by decision tree while bottom classification
method. Likewise, support vector machine, Naive Bayes and decision tree contain
extension associate resulting the bottom techniques on behalf of AdaBoost calculate
in favor of accuracy verification. The AdaBoost contains the accuracy figuring among
decisions tree, i.e., bottom analysis having 90.56%, its very important appeared
another way in relation to that of, decision tree, support vector machine and Naive
Bayes. Artificial awareness is having additional impact is device acknowledging, it
creates estimations arranged to receive in models and decision rule from information.
Artificial intelligence (AI) figuring’s have been entrenched into data mining
channel, it can set them with setup medical methods, to drive out ahead from facts. In
the EU-financed MOSAIC endeavor, a data mining channel has been used to choose a
plan of prophetic models of Category-2 diabetes mellitus (C2DM) [7] traps allowing
for electronic success verification data of accurate approximately 1000 patients. Such
channel incorporates medical center profiling, judicious form focused on, perceptive
model improvement as well as support model. The figured out how to lost records
through techniques on behalf of random forest (RF) which has associated with suit-
able techniques to control the asymmetric classes, we used logistic regression module
decision in the direction of prediction begin the nephropathy, retinopathy, by different
182 D. Bhattacharya et al.
instance conditions, on 5, 7 and 9 years as of the most important visit the hospitals
to check up the diabetes. Measured essentials having sex approach, mature, declara-
tion of time; mass Record (BMI), gluttony hemoglobin, smoking weakness as well
as hypertension. Desire techniques convention built-in according to the complexi-
ties, surrendered a careful to 0.838. Various fundamentals were selected for each
comprehensive nature and time condition, provoking exact models easy to signify
the medical observe.
The article, examine Pima Indian dataset has finished utilizing unusual like clas-
sification measures, logistic regression, Zero R, random forest, Naïve Bayes, J48,
MLP. Examination with expectation of diabetes having the positive or negative.
To test the diabetes then we can use data mining tool, i.e., WEKA tool [8], as far
as correctness and execution MLP is superior. The proposed procedure uses SVM
and an AI strategy as the classifier for examination of diabetes. The AI technique
revolves around organizing diabetes sickness commencing a dataset. The proposed
system is assessed by game plan exactness, k-crease traverse support technique as
well as chaos grid. The required request precision is 93.10% with extraordinarily
accomplished appeared differently in relation to the in advance nitty gritty gathering
techniques.
3 Materials and Methods
3.1 Decision Tree
This is classification technique; this technique is utilized for classification of issues.
The decision tree is a classification technique which divides the two data models from
datasets. The idea of new approach is evaluated for intentional elements. This tech-
nique resolves the informational key and manufactures the prediction of decision
model to the incomprehensible group marks. The classification technique be able
to develop toward equally dual with reliable factors. The decision tree preferably
discovers the reliant root node on its mainly high randomness charge. The deci-
sion tree provides ideal arrangement for selection, and it expects mainly assump-
tion between guidance of dataset. The input of decision tree [9] is set of informa-
tional, comprising an only some qualities and occurrences esteems of the decision
model. problems confronted whereas the decision model structure for choosing from
separation property, parts, uncertain criteria, pruning, preparing test, excellence and
quantity, the request for parts and so forth.
The input is to train the datasets.
The output is to build the decision model for structure of a tree.
The tree structure of a decision model is to incorporate the collection of structure
nodes. The decision nodes incorporate by (divides the form nodes) leaf nodes? The
architecture of decision tree is depicted in Figs. 3, 4 and 5. The dataset having
dissimilar qualities, and the root node has accurate selection of credits to complex
A Novel Approach for Health Analysis Using Machine … 183
job. Each decision node has at least two twigs. The initial node can act as main node
then which is called as root node. This structure identifies the greatest feature because
the initial node otherwise greatest display node through available collection of nodes.
This technique has several approaches for selecting as the finest quality of root node
[10], in the view of levels polluting weight for children’s nodes. This calculates the
performance of classification technique, Gini-index and grouping mistake. These
calculations are accomplished to entire qualities as well as association is completed,
for choosing the optimal spill.
3.2 Naïve Bayesian
The Naïve Bayes is a classification technique, and then it is a feasibility analysis of
technique depends under Bayes theorem [11] shown in the Eqs. (1) and (2), among
self-rule predictions of hypothesis. This technique of dataset can act as input, and it
should perform the analysis with prediction of group label by Bayes Theorem. This
technique measures the possibility of input data into group with the help of compute
the anonymous data samples in the group. This technique is used for appropriate of
huge datasets. The given below formula is a Naïve Bayes formula, which is used
for calculation of posterior probability of all groups. The Naïve Bayes technique of
flowchart is shown below in Fig. 1.
Q(b|y) = Q(b|y)Q(b)
Q(y) (1)
Q(b|y) = Q(y1|b) ∗ Q(y2|b) ∗ .... ∗ Q(yn|b) ∗ Q(b)(2)
Q(b|y) which contains the group of posterior probability (goal) known analyst
(element).
Q(b) which contains the group of prior probability.
Q(y|b) which contains the possibility of analyst probability in known group.
Q(y) which contains the analyst for prior probability. Support vector machine.
The support vector machine is a classification algorithm, selective arrangement
approach. The approach is used in favor of jointly classification as well as regression.
This justification is done among the datasets after discovery of SVM has manic line,
this technique can be partitioned into two classes of best datasets depicted in Fig. 2.
This incorporates the two stages, the observation of benefits otherwise perfect manic
line during information gap along with the restrictions determined by the mapping
of objects. This technique constructs the representation of model, which allocates
for latest example classes.
184 D. Bhattacharya et al.
Fig. 2 Data allocation of
support vector machine
under manic line
3.3 Random Forest
The random forest is a classification algorithm, this algorithm mostly used for clas-
sification problems. This supervised learning algorithm also used for together clas-
sification as well as regression. This justification of the decision manic line has
rear of the support vector machine into datasets, and this algorithm of the datasets
can be divided into two classes is depicted into Fig. 5. This incorporates the two
stages, the observation of benefits otherwise perfect manic line during information
gap along with the restrictions determined by the mapping of objects. This support
vector machine constructs the representation of model, which allocates for l atest
example classes.
Weight the information where it comprises “m” highlights talking to the behavior
of the dataset.
The preparing computation of asymmetrical random forest is known as bootstrap
calculation otherwise stowing method randomly to select the—n highlight from—
m highlights, for example, to build the arbitrary patterns, the classification technique
directs the innovative examples of patterns selects the OOB fault.
The Determine of node—d utilizing greatest divide. Mainly, the sub-nodes
divided by the main nodes.
The meaning of Repeat is, to find out the—n number of trees.
The random forest technique makes the decision trees on dataset samples along
with it takes the prediction from each of them and lastly choose the top answer
through resources of selection.
3.4 K-Nearest Neighbor (KNN)
Customized Method.
The altered methodology incorporates the purpose of the accurate properties from
the huge information support, in the clarification of dataset problems affected by the
classification problems. Mainly, each problem contains the accurate/perfect behavior,
A Novel Approach for Health Analysis Using Machine … 185
Table 1 Outcomes for
support vector machine S. No. Characteristics Explanation
1 Length of life It is a time of human life
2Sex Female or male
3Clot lactose diet –
4 Clot lactose position
prance
–
5Carrying Carrying add up for ladies
6level of Blood lactose To test the glucose level in
the blood
this obtains the individual analysis of overlooking for dispensable properties. These
datasets of information depicted into Table-1 incorporates a variety of properties
and its illustration. Purpose of the exact credits sticks to the feature information
dataset and feature outcomes comes by grouping could be typical. This methodology
incorporates five stages.
This type of problems will be expressed by qualities as well as properties.
Transmission and dataset collection that credits come by mj = 0 and ml =
maximum, which means maximum is nothing but a no. of properties; moreover,
J has quality is—one (central driver).
Representation of diabetes: intensity of sugar qualities expresses the type of
persons experiencing diabetes.
This Attribute has value is 1, then it is considered as input (the explanation of
major properties is accountable).
This Attribute has value is 1, this is known as process provides the relation for
additional feature m, by the values has been produced.
The output is that the attributes of selection, and this type of classification
outcomes could be enhanced (Table 1).
Correlation value = ⎡
Attribute ax −
n
Σ
i=1
Attribute(xi ) ⎤
− 1
The method is proceeded with various properties, principles are contrasted and
every feature (attribute), on the off chance that the value dissimilarity is more than
the new attribute, at that position feature has fewer importance, for example regard 1
is contrasted and regard n. The top uniqueness is selected and masterminded in a vast
demand, and the previous model highlights the datasets that is used for classification
events.
186 D. Bhattacharya et al.
4 Results and Discussions
This appearance review for classify systems be prepared during different completing
procedures, for example, accuracy, affectability, distinctiveness, correctness and
analysis. Our study article center about the five collections of measures, for example,
decision tree, SVM, Naïve Bayesian, KNN as well as random forest. Table 2 depicts
for outcomes that allocated to supervised learning method. Our testing is lead during
immediate digger classification tools.
SVM: This supervised learning method be functional happening clinical datasets.
This classification method has the accuracy i s 61.45%. These outcomes are displayed
into Table 3.
The Random Forest: This classification method has the accuracy is 68.46%.
These outcomes are displayed into Table 2. The illustration of the tree structure
is distinctiveness depends on dissimilar circumstances into Fig. 7.
The Classification of Naive Bayes’: This correctness be 70.67%. These outcomes
are displayed into Table 3.
The classification of Decision tree: This accurateness of value is 82.34%. These
outcomes are displayed into Tables 1 and 2. This three-picture displayed into Fig. 2.
Table 2 Method of classification results
S. No The Method of
classification
Accurateness Appropriately classify Inaccurately classify
1Naïve Bayes 70.67 218 72
2 Decision tree 82.34 653 315
3Support vector
machine
61.45 685 262
4Random forest 68.46 688 276
5K-nearest
neighbor
79.41 232 94
Table 3 Outcomes for
support vector machine Accurateness =
61.45
Non-diabetic of
accurate
Diabetic of
accurate
Group
accuracy
Predefined of
non-diabetic
268 65 81.34
Predefined of
diabetic
223 333 80.67
group remind 67.34 90.72
A Novel Approach for Health Analysis Using Machine … 187
5 Conclusion and Future Scope
It symbolizes the meaning of diabetes is a miscellaneous (various) collection of
illnesses. This is represented by the fact that blood contains glucose. The primary
goal of the American Diabetes Association is “to prevent and secure diabetes and to
develop lives that are astonishingly unaffected by diabetes.” The primary assistance
for individual lives all over the world, to try to classify as well as stop the inconve-
niences for diabetes at the start of the prescient anal-sister can be used to improve
their arrangement systems. The proposed system also performs the investigation for
highlight datasets as well as the selection of ideal highlights based on reputable
relationships and relational relationships. The two algorithms with the highest accu-
racy values are random forest and decision tree, which have 99.11% and 99.62%,
respectively. Individual investigation is the best technique for collecting diabetic
information. The Naive Bayes’ and support vector machine procedures provide that
the exactness of values is 90.34% and 76.91%, respectively, by current strategy. As a
result, the proposed technique is used to improve the precision of grouping systems.
The accuracy of the improved support vector machine is 82.53%, and the precision
of the Naive Bayes is 81.59%; thus, this technique can be defined as beginning with
low dimensions and ending with high measurements were successfully obtained.
This provides accurate information for the patient’s records of both diabetic and
non-diabetic patients’ data. So that the disease’s frequency rate can be predicted and
the most.
References
1. Global Report on Diabetes 2016 by World Health Organisation. ISBN 978-924-1565257. http://
www.who.int/diabetes/publications/grd-2016/en/
2. Chandra Sekhar P, Thirupathi Rao N, Bhattacharyya D, Kim T (2021) Segmentation of natural
images with k-means and hierarchical algorithm based on mixture of Pearson distributions. J
Sci Ind Res 80(8):707–715. Retrieved from www.scopus.com
3. Yıldırım Ö, Pławiak P, Tan RS, Acharya UR (2018) Arrhythmia detection using deep
convolutional neural network with long duration ECG signals. Comput Biol Med 102:411–420
4. Uddin MZ, Dysthe KK, Følstad A et al (2022) Deep learning for prediction of depressive
symptoms in a large textual dataset. Neural Comput Appl 34:721–744. https://doi.org/10.1007/
s00521-021-06426-4
5. Mandhala VN, Bhattacharyya D, Vamsi B, Thirupathi Rao N (2020) Object detection using
machine learning for visually impaired people. Int J Current Res Rev 12(20):157–167. https://
doi.org/10.31782/IJCRR.2020.122032
6. Bhattacharyya D, Kumari NMJ, Joshua ESN, Rao NT (2020) Advanced empirical studies on
group governance of the novel corona virus, Mers, Sars and Ebola: a systematic study. Int J
Current Res Rev 12(18):35–41. https://doi.org/10.31782/IJCRR.2020.121828
7. Donahue J et al (2015) Long-term recurrent convolutional networks for visual recognition and
description. In: Proceedings of the IEEE conference on computer vision and pattern recognition
(CVPR)
188 D. Bhattacharya et al.
8. Eali SNJ, Bhattacharyya D, Nakka TR, Hong S (2022) A novel approach in bio-medical image
segmentation for analyzing brain cancer images with U-NET semantic segmentation and TPLD
models using SVM. Traitement Signal 39(2):419–430. https://doi.org/10.18280/ts.390203
9. Bhattacharyya D, Doppala BP, Thirupathi Rao N (2020) Prediction and forecasting of persistent
kidney problems using machine learning algorithms. Int J Current Res Rev 12(20):134–139.
https://doi.org/10.31782/IJCRR.2020.122031
10. Um TT et al (2017) Data augmentation of wearable sensor data for Parkinson’s disease moni-
toring using convolutional neural networks. In: Proceedings of the 19th ACM International
conference on multimodal interaction, pp 216–220
11. Zhai J, Barreto A (2006) Stress recognition using non-invasive technology. In: Proceedings of
the 19th International Florida artificial intelligence research society conference FLAIRS, pp
395–400
Classification of Healthy and Diseased
Lungs by Pneumonia Using X-Rays
and Gene Sequencing With Deep
Learning Approaches
Debdatta Bhattacharya , K. V. Satyanarayana, N. Thirupathi Rao ,
and Eali Stephen Neal Joshua
Abstract This research work is entitled to predict the lung disease using chest X-rays
by deep learning technique. Lung disease is a term that refers to improper functioning
of lungs. There are many diseases which occur due to the abnormal functioning of
lungs. It includes tuberculosis, pneumonia, lung cancer, and asthma. The infection
can be bacterial, viral, or fungal. It causes inflammation of trachea and respiratory
failure. If found earlier, it can be cured, or else it can even lead to death. This project
classifies the normal and abnormal X-ray with a percentage of accuracy so that we
can give the treatment to the patient accordingly by seeing the X-ray. Algorithms
used are convolutional neural network (CNN) and Inception Neural Network (INN)
and TensorFlow which is Google open-source algorithm. The project is helpful for
finding lung disease using chest X-ray.
Keywords Convolutional neural networks ·Inception v3 model ·Inception neural
network ·Trachea ·Pneumonia ·Bronchitis
1 Introduction
When the lungs are unable to perform as effectively as they should because of a
disease or condition, this is referred to as a sickness of the lungs. Breathing problems
are almost often brought on by some kind of lung disease. There are more than 32
different forms of lung issues, some of which include asthma, bronchitis, chronic
D. Bhattacharya (B)
Department of Computer Science and Engineering, Koneru Laksmaiah Education Foundation,
Vaddeswaram, Guntur, Andhra Pradesh 522302, India
e-mail: debdatta122001@gmail.com
K. V. Satyanarayana
Department of Computer Science and Engineering, Raghu Engineering College, Visakhapatnam,
Andhra Pradesh, India
N. T. Rao · E. S. N. Joshua
Department of Computer Science and Engineering, Vignan’s Institute of Information Technology,
Visakhapatnam, Andhra Pradesh 530016, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_20
189
190 D. Bhattacharya et al.
illness, influenza, pneumonia, and others. Lung diseases are the third most common
reason people pass away anywhere in the globe. In India, lung illness is the main
cause of mortality in children under the age of one. Because of that, at least one
person loses their life every single day. Patients who have this problem identified
early on have a better chance of avoiding its long-term effects. We used the Inception
v3 model to evaluate the state of the lungs and come to the conclusion that they were
healthy. Convolutional neural network (CNN) and Inception Neural Network (INN)
are all various names for the same method that is often used to categorize pictures
into distinct groups. This method is known as the Inception Neural Network (INN).
When trying to classify photographs, deep learning algorithms often turn to the
use of convolutional neural networks. In order to give each layer its own distinc-
tive appearance, filters are applied to them. A convolutional neural network has
three layers: the convolutional layer, the pooling layer, and the fully connected layer.
The convolutional layer is the topmost layer. These levels include the classification-
specific filters, if you are looking for them. As part of this experiment, the researchers
are comparing pictures obtained from a lung that has been infected with those
obtained from a lung that has not been infected. It is possible to make compar-
isons based on the color of the lungs while sorting. There are several viruses and
bacteria that have been linked to the development of lung illness. Those who are
healthy will have lungs that are a different color from those who have respiratory
illnesses such as pneumonia, bronchitis, or coronavirus. Setting up the filter with
the training picture is required in order to find the pulmonary walls in the image.
There are two distinct kinds of pooling that take place in the pooling layer: maximal
and average. Pooling determines an average of the number of things that have been
filtered out, while maximum pooling chooses the greatest number of items that have
been filtered out.
When an infection is diagnosed and treated at an early enough stage, there is a
chance that lung failure may be avoided. When utilized together, the convolutional
neural network and the Inception Neural Network will make it much easier to deter-
mine whether or not a patient’s lungs have been contaminated by germs or viruses.
Illness brought on by antibiotics in order to stop the model from coming up with
inaccurate predictions, it has to be trained to have the greatest possible level of preci-
sion. The code for this project was developed with the assistance of already existing
systems, numerous different pieces of study on long illness, the convolution neural
network, and the inception model. This project, which will ultimately broaden its
scope to include the identification of damaged organs, will make use of scanned
photos as its dataset. Using scan photographs like these as input, it is possible to
determine if an organ is healthy or ill.
2 Literature Survey
In 1962, Hu introduces algebraic entropy. David Hilbert [1] may have affected him.
Moment invariants were first employed to solve community problems. They were
Classification of Healthy and Diseased Lungs by Pneumonia … 191
compared with algebraic invariants and his seven two-dimensional invariants. PDSs
are low-level encoding schemes. These systems take picture data and produce an
image with noise reduction and sharpening. The original image is utilized to make an
intermediate copy, which is used to recognize and isolate items. The original picture
is used. Uppal Uri and colleagues [2] developed an extended territory-locating algo-
rithm. Using 15 statistical and fractal texture characteristics, this approach divides
tiny regions into six categories. Radiologists must first identify the two lungs in an
X-ray picture before looking for issues. CAD systems concentrate on one body area,
such as the thorax, breast, or colon, and employ X-rays, CT scans, PET scans, and
MRI (MRI).
Gabor filtering was proposed by Manish Kakar and others [3] to recover picture
texture information. The segmentation’s delineation accuracy was above 90%.
Combining form position-based data with cortex-like function increased simple SVM
classifier sensitivity to 89.48%. Automatic segmentation did this. Summers said this
procedure takes a long time and there are too many photos. This is important in
nations with many populations but few physicians. Kim Ko and Jung solved DFR
using neural networks and time intervals. A CT scan is usually more informative
and accurate than an X-ray. Doctors rely on chest X-rays to identify lung cancer and
TB early. F.-Y. Zou [4] suggested two ways. The first modifies wavelets to minimize
image noise, while the second evaluates edge detection operators like Differential,
Log, Canny, and Binary morphology. Both boost picture quality. Based on simula-
tion findings, positive and negative characteristics of several edge detection operators
were studied. Binary morphology may enhance edge appearance. The borders-closed
approach was proposed as a last-ditch attempt to acquire a full image profile. Abby A.
Goodrum [5] writes “Image information retrieval: a survey of current research” These
include: Text-based, content-based, and image retrieval user experiences. [Book]
This comprehensive method uses several domains. The paper “Feature Selection:
Evaluation, Application, and Small Sample Performance” [6] by Pudil et al. shows
that Jain and A. D. Zinger’s solution outperforms other approaches. We utilize SAR
satellite images and four texturing models to find a set of features that may be used to
define land uses. When characteristics from several texture models and other factors
are added in categorization, accuracy improves.
3 Materials and Methods
3.1 Convolutional Neural Network (CNN)
A particular type of feed-forward artificial neural network influenced by the visual
cortex is the Coevolutionary Neural Network. The visual cortex is little more than a
tiny area that is sensitive to particular areas of the visual field in our brain, helping us
to identify things viewed by us. In the Coevolutionary Neural Network, the neuron
in a layer will be connected only to a small region of the layer before it, instead of all
192 D. Bhattacharya et al.
Fig. 1 Showing CNN architecture with improvised Max-pool layers
the neurons; [7] in a fully connected way as used in fully connected networks. The
following layers are compatible with CNN that is the convolution layer, the pooling
layer, the ReLu [8] layer, and lastly the completely linked layer.
The convolutional neural network consists of three layers in which each layer has
filters which are specific to it. The pooling layer will get its input from the previous
convolution layer. There are two types of pooling layers, namely max pooling [9]
and average pooling. The max pooling layer selects the maximum elements in the
areas covered by the filter, whereas the average pooling layer selects only the average
number of elements in the area covered by the filter. The output from the pooling
layer is passed as the input to the next layer called fully connected neural layer. After
going through all t he filters in each layer, the input image from the dataset will get
classified based on the s pecification of the filter. Once the machine gets trained for
the input dataset, the newly incoming images can be classified easily by the model
with high accuracy (Fig. 1).
3.2 Inception V3 Model
Inception models contain two important parts: the fully extracted part of convo-
lutional networks and the classified part of fully connected networks. In the first
part, the typical features of images are extracted from the input, while in the rest,
images are classified based on their features. Early v3 models are pre-trained deep
learning models that achieve state-of-the-art precision in identifying general objects.
It contains many layers and many networks, and at each layer, features are extracted
and stored for classification.
Classification of Healthy and Diseased Lungs by Pneumonia … 193
Fig. 2 Showing the model of the proposed inception V3 model
4 Proposed Model
The whole algorithm accepts a Keras object classifier model [10], which can be loaded
with post ImageNet weights if desired. For concise examples of image classification
use cases, see this research. Be sure to read the guide to transfer learning and fine-
tuning for transfer learning use cases. Recognize that each Keras Application needs
a different type of input preprocessing. Call tf. keras applications for Inception V3.
Until forwarding the inputs to the model, use inception v3 preprocess input. Input
pixels will be scaled between −1 and 1 by inception v3 preprocess input (Fig. 2).
5 Results and Discussions
The initial photos of reactive lymphoid hyperplasia, NHL, SCC, and adenocarcinoma
had classification accuracies of 88.46%, 80.77%, 89.29%, and 100%, respectively.
On the test dataset, the overall accuracy was 99.62%. The uncertainty medium of
the organization Cohen’s kappa [8] was used to measure the agreement between
cytopathologic and DCNN which was 0.8620.077. Three disjointed imageries of
RDH and three split descriptions of SCCs were incorrectly labeled as NHLs.
We dug deeper into the untreatable photos to figure out why they failed. The
disjointed images of sensitive lymphoid hyperplasia that were misdiagnosed as NHL
are shown in Fig. 3. The disjointed photos of NHL [11] that were misdiagnosed as
reactive lymphoid hyperplasia. The disjointed photos of NHL that were misdiagnosed
as SCC and adenocarcinoma are shown in Fig. 3. The fragmented images of SCC
that were misdiagnosed as NHL are shown in Fig. 2. The cytopathologic’ analyses
of the images are represented in the figure legends.
194 D. Bhattacharya et al.
Fig. 3 Showing the accuracy of the Inception V3 Model with comparison of CNN architecture
6 Proof of Concept
6.1 POC Architecture
In the proposed model, more than 4000 images were provided for training purposes.
The model uses convolutional neural networks to extract many features for each
image. For each feature, a matrix of vector values was assigned. While passing
through the model, the matrix values decrease and finally a series of values remain.
The model stored array values for all tagged images. Given an unknown image, we
apply the same method to extract the features and then compare the extracted features
with the stored features of the training image using a fully connected network. Results
are displayed based on accuracy. This model has the highest accuracy of 92.57% so
far. The user’s input image is being entered into the model. After classification, the
image is displayed in a separate window with the corresponding results.
6.2 POC Design
For more user-friendly purposes, a graphical user interface has been developed to
include images in the model. The model can run on multiple images as well as a
single image, allowing users to clearly identify diseases in a single plant using many
images of a particular plant from different angles.
This GUI was developed using Python packages such as Tkinter and Bellow.
Images containing classification results are displayed in a separate window,
improving efficiency.
Classification of Healthy and Diseased Lungs by Pneumonia … 195
6.3 Result
With the help of the proposed model with 92.57% prediction accuracy, users can
easily detect lung images and classify them as normal or abnormal, helping patients
receive treatment in advance. This helps doctors see the severity of the patient and
treat it accordingly.
7 Conclusion and Future Work
The approach presented here makes it easier for patients to identify normal or
abnormal lung scan findings, with a 92.57% accuracy rate. Patients can decide
whether they require therapy before it is too late. Patients might be treated reason-
ably, given their illness. Through training on their dataset, scientists built a model
that could distinguish lung disease and other normal or aberrant images that might
affect organs. This model did not only identify lung disease. Any bodily part might
be affected by sickness. This method improves mental health patients’ quality of life.
This method helps people pick and implement solutions for solvable problems.
References
1. Xie Y, Xia Y, Zhang J, Feng DD, Fulham M, Cai W (2017) Transferable multi-model ensemble
for benign-malignant lung nodule classification on chest CTBT—medical image computing
and computer-assisted intervention—MICCAI 2017, pp 656–664
2. Cao H, Liu H, Song E, Ma G, Xu X, Jin R, Liu T, Hung C (2019) Multi-branch ensemble
learning architecture based on 3D CNN for false positive reduction in lung nodule detection.
IEEE Access 7:67380–67391
3. Bhattacharyya D, Dinesh Reddy B, Kumari NMJ, Rao NT (2021) Comprehensive analysis on
comparison of machine learning and deep learning applications on cardiac arrest. J Med Pharm
Allied Sci 10(4):3125–3131. https://doi.org/10.22270/jmpas.V10I4.1395
4. Armato SG 3rd et al (2011) The lung image database consortium (LIDC) and image database
resource initiative (IDRI): a completed reference database of lung nodules on CT scans. Med
Phys 38(2):915–931. https://doi.org/10.1118/1.3528204
5. Han F, Wang H, Zhang G, Han H, Song B, Li L, Moore W, Lu H, Zhao H, Liang Z (2014)
Texture feature analysis for computer-aided diagnosis on pulmonary nodules. J Digit Imaging
28:99–115
6. Wei G, Ma H, Qian W, Han F, Jiang H, Qi S, Qiu M (2018) Lung nodule classification using
local kernel regression models with out-of-sample extension. Biomed Signal Process Control
40:1–9
7. Shen W, Zhou M, Yang F, Yu D, Dong D, Yang C, Zang Y, Tian J (2016) Multi-crop convo-
lutional neural networks for lung nodule malignancy suspiciousness classification. Pattern
Recognit 61
8. Eali SNJ, Bhattacharyya D, Nakka TR, Hong S (2022) A novel approach in bio-medical image
segmentation for analyzing brain cancer images with U-NET semantic segmentation and TPLD
models using SVM. Traitement Signal 39(2):419–430. https://doi.org/10.18280/ts.390203
196 D. Bhattacharya et al.
9. Bhattacharyya D, Doppala BP, Thirupathi Rao N (2020) Prediction and forecasting of persistent
kidney problems using machine learning algorithms. Int J Current Res Rev 12(20):134–139.
https://doi.org/10.31782/IJCRR.2020.122031
10. Togacar M, Ergen B, Cömert Z (2019) Detection of lung cancer on chest CT images using
minimum redundancy maximum relevance feature selection method with convolutional neural
networks. Biocybern Biomed Eng 40
11. Thakur SK, Singh DP, Choudhary J (2022) Lung cancer: detection and classification of malig-
nancies. In: Dubey HM, Pandit M, Srivastava L, Panigrahi BK (eds) Artificial intelligence and
sustainable computing. algorithms for intelligent systems. Springer, Singapore. https://doi.org/
10.1007/978-981-16-1220-6_38
Breast Cancer Classification Using
Improved Fuzzy C-Means Algorithm
N. Thirupathi Rao , K. V. Satyanarayana, M. Satyanarayana,
Eali Stephen Neal Joshua , and Debnath Bhattacharyya
Abstract Abnormal growth in the breast tissue prompts to the strange cell devel-
opment in the breast. To decipher this statement in a mammogram precisely, the
quality of the pictures ought to be at its incomparable. The proposed research work
is conveyed out for examinations of different screening strategies to recognize the
unique phases of breast malignancy. In India for every 4 min, the women are diag-
nosed with this disease. And a woman dies with this disease for every 13 min. This
disease is prominent with the people living in the ruler area while comparing the
people in the urban areas. Therefore, it is very important to find and treat this disease
as early as possible. The breast tumor region, perimeter and breadth are assessed from
mammogram picture databases. The Bits Errors Degree (BER), Highest Indication
to Clatter Percentage (PSNR) and Callous Tetragonal Inaccuracy (MSE) values are
determined for both abnormal and normal images. These analyses were used to
approve the presence or absence of the disease and to support the evaluation process
for finding the disease. This quality assessment is used to understand the reality on
Earth for a specific diagnosis that is a specific type of chromatin in a carcinogenic
core that may indicate an irregular protein sequence.
Keywords PSNR ·MSE ·Malignancy ·Cancer ·Classification ·Fuzzy means
algorithm
N. T. Rao (B) · E. S. N. Joshua
Department of Computer Science and Engineering, Vignan’s Institute of Information Technology,
Visakhapatnam, Andhra Pradesh 530016, India
e-mail: nakkathiru@gmail.com
K. V. Satyanarayana · M. Satyanarayana
Department of Computer Science and Engineering, Raghu Engineering College, Visakhapatnam,
Andhra Pradesh, India
D. Bhattacharyya
Department of Computer Science and Engineering, Koneru Laksmaiah Education Foundation,
Vaddeswaram, Guntur, Andhra Pradesh 522302, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_21
197
198 N. T. Rao et al.
1 Introduction
This article explains about the importance of breast cancer [1]. Breast cancer accounts
for 28–35% of all lady malignancies in all conurbations transversely India. It is very
important to find the breast cancer at earlier stage. All over the country, U.S is mostly
affected with this disease. In India, for every 4 min [2] the women are diagnosed
with this disease. And a woman dies with this disease for every 13 min. This disease
is more common with the people living in rural area [3] while comparing the people
in urban area. Therefore, it is very important to find and treat this disease as early as
possible.
According to the affected areas and severeness of the problem [4], this cancer
is divided into different stages; with the different stage, it is divided into different
types; and the early-stage cancer is more treatable while comparing the later stages.
Motivation and Previous Work.
Athertya et al. (2016) developed an involuntary separation of delineations from
CT images using fuzzy crooks [5]. In this method, the automatic initialization of
contours has been demonstrated using active contour method. Fuzzy corner gave
an accuracy of 80% with high Dice coefficient and low Hausdorff distance. This
algorithm is suitable for noisy images also. It might be a daunting task in case of lax
tissue image means, and it has the complexity in finding of corners of the image. Elisee
Ilunga Mbuyamba et al. (2016) proposed an alternative active contour model (ACM)
driven by Multi-population Cuckoo Search algorithm [6]. This strategy assists the
converging of control points toward the global [7] minimum of the energy function
unlike ACM which is often trapped in local minimum. The algorithm has been tested
and implemented on MRI images. Three metrics, Jaccard index, Dice coefficient and
Hausdorff distances [8], have been used to assess the results. This method requires
lesser iterations and is robust and more effective. It takes more computational time
for computing the iterations. Agus Pratondo et al. (2016) delivered improved robust
Edge Stop Functions (ESFs) for edge-based active contour models [2]. Robust Edge
Stop functions use gradient information which fails to stop contour evolution when
the image has poor boundaries. In this method, the new ESFs have been used which
have gradient information as well as probability scores to classify the mass. This
method was evaluated using two quantitative measures, namely Jaccard index and
Dice coefficient. This method converges faster and gives global contours but it is
a complex method. Radha et al. (2016) proposed an image enhancement technique
for breast cancer detection [9]. Mean filters, median filter, Wiener filter and linear
filter are used for pre-processing; among these filters, median filter provides best
results. Image segmentation is performed through thresholding technique and K-
means algorithm. The tumor edges are detected using canny edge detection technique.
This algorithm results in a higher accuracy. The limitation of this method is the
difficulty in finding blurred image edges.
Breast Cancer Classification Using Improved Fuzzy … 199
Fig. 1 Diagrammatic flow of the proposed method
2 Proposed Work
The proposed method comprises four steps. In the first step, pre-processing is done
where the unwanted parts such as labels have been removed. In the second step,
optimization has been done where the image gets optimized for the further processing
methods. In the third step, segmentation has been done where the exact affected parts
can be obtained. Segmentation is followed up by the fourth step feature extraction
where some special features get extracted and made ready for the classification.
2.1 Pre-processing
The main purpose of pre-processing is to improve the image quality in an effective
manner. The proposed method consists of few pre-processing steps; in the first step,
background image should get removed; in the second step, pectoral muscle should
get removed based on the image orientation; and in the third step, the image should
get enhanced where the quality should get improved without any artifacts. Figure 1
shows the flow diagram of pre-processing method.
2.2 Labels and Other Artifacts Removal from the Background
To find the tissue boundary of the breast, we do the following: First, we switch
from the unsigned portion of the image to the double decade, and then, we get the
image energy, which is equal to the second power of the decimal image. The model
of image energy is shown in figure after conversion; once again according to the
200 N. T. Rao et al.
threshold value, we transform that again to the original binary image. Background
areas will not cover the breast area, and it mostly occurs in the right and left of the
image. As the background regions are dark and their gray level value will always be
close to zero, this gray level does not differ from each other. Therefore, find the new
function by fixing an intensity value. That is, here the taken parameters are fixed with
the specified value; the range within that fixed value is given a maximum intensity
value. The range above and below the fixed specified value is denoted.
2.3 Removing of Pectoral Muscle
The pectoral muscle is a muscle that comes behind the border of the breast. This is
an unwanted portion that basically comes in the mammogram image; therefore, the
removal of this region will be helpful for the further segmentation process. There
are various methods available to remove that region. In this thesis, we remove that
unwanted region by finding the correct border of the breast part. By finding the
correct border, it is easy to evacuate the remaining unwanted regions.
2.4 Enhancement of Image
Enhancement is the final step in the pre-processing techniques. During this enhance-
ment, the quality of the image gets improved, which is more important for the next
undergoing process. Spatial domain and frequency domain are the two basic classi-
fication techniques of image enhancement in our proposed method; median filtering
is used to enhance image equality. Median filtering and enhancement are basically
done by calculating the median value of the image pixel value. Algorithm to find the
median value is shown below.
Step 1: The pectoral muscle removed image is obtained
Step 2: If the obtained pixel is noisy, it should undergo a further process
Step 3: Replace all the noisy pixels using the median value
Step 4: Shift the window
Step 5: Repeat step 3 for all the pixel values
Step 6: Obtain the enhanced image.
2.5 Measurement of PSNR Value and MSE Value
The performance of this median filter is calculated using the peak signal-to-noise
ratio (PSNR) value and MSE values.
Breast Cancer Classification Using Improved Fuzzy … 201
(x + a)n =
n
k=0n
kxk an−k (1)
Peak signal-to-noise ratio (PSNR) is the ratio between the maximum potential
value of a signal (power) and the noise distortion that affects the quality of its repre-
sentation. Since many signals have a very wide dynamic range (the ratio between the
largest and smallest values of the convertible size), the PSNR is usually expressed in
terms of logarithmic decibels. Improving the image or improving the visual quality
of a digital image is subjective. Here, the obtained PSNR value is 35.28 which shows
that the final enhanced image is better in quality. MSE value is 2.9861 which shows
the result has less error and the obtained image quality is very good. In contrast
to the proposed technique gives more accurate result. These results suggest that
our current study has convincingly enhanced the quality of the image with better
in contrast saying that a system provides a high-quality picture is different for each
person. For this reason, it is necessary to establish quantitative/empirical measures to
compare the effects of image enhancement methods on image quality. Using similar
test images, we can systematically compare different image enhancement methods.
To identify if a particular method produces better results. The metric under investi-
gation is the peak-to-signal-to-noise ratio. If we can show that an algorithm that is
similar to the original or a set of instructions can improve the known image of the
degenerate, then we can more accurately conclude that this is a better algorithm.
3 Optimization Segmentation
Basically, the medical images are not very crystal clear, it contains more noises, and
picture quality is also very low while comparing other digital images, so segmenting
those images directly may lead to poor segmentation; therefore, the identification
of cancerous cells becomes complex. To reduce those complexities, the obtained
medical image is first optimized.
3.1 Optimization Using Independent Search Krill Herd
Technique
In this manuscript, a brilliant algorithm, named krill herd (KH), is used t o solve the
optimization tasks successfully. The KH algorithm works according to the simulation
behavior of the krill members. The krill members try to preserve a more density which
moves automatically due to their mutual effects. For a krill individual, this movement
can be defined as:
202 N. T. Rao et al.
f (x) = a0 +
∞
n=1an cos nπ x
L + bn sin nπ x
L(2)
Nmax is denoted as the maximum speed and on is denoted as the weight of the
motion induced in the range [0, 1], N old is the last motion induced provide due to
the fellow citizen, and a target j is the objective route effect providing by the best
rill distinct. The measured values of the maximum induced speed are considered as
0.02(MS). The result of the neighbors can be considered as an attractive/repulsive
tendency among the individuals for a neighborhood search. In this study, the result
of the neighbors in a krill movement member is strong-minded as surveys:
sin α ± sin β = 2sin 1
2 (α ± β)
cos 1
2 (α ∓ β)(3)
The neighbor’s vector might be appealing or appalling since the standardized
esteem be negative or positive. The distance for every krill individual can be resolved
utilizing diverse techniques. Here, it is resolved by utilizing accompanying equation
for every cycle:
cos α + cos ds(β) = 2 cos 1
2 (α + β)
cos 1
2 (α − β)(4)
where ds, i denotes the distance for ith krill individual and N denotes the total
number of krill individuals, and the value 5 given in the denominator of the equation
is empirically found. By utilizing the above condition, if the separation of two krill
individual is not exactly the characterized distance, then they are neighbors. The
well-known main vector of every krill member is the most reduced fitness member.
The krill member with the best fitness on the ith singular krill is considered utilizing
the above equation.
Where, Cbest is the most successful member of the krill movement with the best
wellness to the ith krill membership. The Cbest is characterized as an objective which
drives the answer for the worldwide optima, and it ought to be more powerful than
other krill individual, such as neighbors.
In Fig. 2, it is shown that the first graph produces the fitness curve output using
independent free search krill herd optimization technique. The second graph shows
the fitness curve output using PSO technique, and from the graph, it is clear that
IFSKHO technique produces best fitness value while comparing PSO technique.
4 Results and Discussions
Classification is a process used in medical image processing to distinguish benign and
malignant tumor cells. Breast cancer classifies according to different program criteria
and serves a different purpose. The major categories are histopathological type, tumor
Breast Cancer Classification Using Improved Fuzzy … 203
Fig. 2 Fitness curve, a
fitness curve output of
proposed method and b
fitness curve output of PSO
technique
quality, t umor status and expression of proteins and genes. In this thesis, the Gray
Level Co-occurrence Matrix is used to distinguish the cancerous and non-cancerous
cells. They are the size of the tumor, mobility, their spread in the lymph nodes and
their spread in the other parts. Basically, the work of classifiers is to classify the good
tissue and the bad tissue, which is said in other words that classifier has to classify the
cancer cells and non-cancerous cells. The classification of cancer cells can be done by
identifying the subtype count in the feature extracted cells. Estrogen Receptor (ER),
Human Epidermal Growth Factor (HER2), basal-like luminal-A and luminal-B are
most probably used subtypes as a predictive factor for classifying abnormal cells
from the normal cells. Our analyses were led in two sections, to assess these marks
(starting now and into the foreseeable future we allude to every single prognostic
signature and organic pathways as just marks, except if explicitly recognized) for
their capacity. Breast cancers are classified into known sub-atomic or subtypes (basal-
like, HER2-enhanced, luminal-A and luminal-B) and ER status (ER+ and ER−). The
figure shows the various subtype occurrence in the different stages of cancer and the
methods used currently to treat the cancer cells. AR stands for Androgen Receptor,
BCL is B cell Lymphoma, CK is Cytokeratin, EGRF is Epidermal Growth Factor
Receptor, ER is Estrogen Receptor, ERBB2 is Erythroblastic Leukemia Oncogene
homolog, PARP is poly (ADB-ribose) polymerases, and PR is Progesterone Receptor.
Appraisals of mammography sensitivity can run from 75 to 90% with specificity from
90 to 95%. The positive prescient estimation of mammography for bosom malignant
growth ranges from 20% in ladies under age 50–60% to 80% in ladies aged 50–69.
And the negative prescient estimation can range from 90 to 95% in the age > 40 years
in women (Table 1).
204 N. T. Rao et al.
Table 1 The values of
reliability ratio with the
diagnostic procedure
Parameters Diagnostic procedures
Mammography
Sensitivity (%) 52.4
Specificity (%) 66.7
Positive predictive value (%) 40.7
Negative predictive value (%) 76.2
5 Conclusion and Future Work
We developed a new methodology for automatic mass detection to aid in the manual
identification of masses in mammography images. The suggested project begins by
identifying the t umor inside the specified image region of interest using a fuzzy c-
means technique and then verifies the image features (i.e., texture) produced from
the FCM input data using the GLCM feature texture to help in the segmentation
process. As can be observed, the suggested methodology’s results were more closely
matched with those from the Mini-MIAS database, demonstrating that the suggested
technique is capable of properly and automatically extracting masses from ROI.
References
1. International Agency for Research on Cancer (2018) Latest global cancer data: cancer
burden rises to 18.1 million new cases and 9.6 million cancer deaths in 2018 [Press
release]. 12 September. Available at: https://www.iarc.fr/wp-content/uploads/2018/09/pr263_
E.pdf. Accessed 21 Feb 2022
2. Parvathavarthini S, Visalakshi NK, Shanthi S, Mohan JM (2018) Crow search optimization
based fuzzy c-means clustering for optimal centroid initialization. TAGA J Graph Technol
14:3034–3045
3. Ravnik J, Jovanovac A, Trupej N, Vistica M (2021) A sigmoid regression and artificial neural
network models for day-ahead natural gas usage forecasting. Cleaner Responsible Consumption
3. Article 100040
4. Bhattacharyya D, Doppala BP, Thirupathi Rao N (2020) Prediction and forecasting of persistent
kidney problems using machine learning algorithms. Int J Current Res Rev 12(20):134–139.
https://doi.org/10.31782/IJCRR.2020.122031
5. Sable SD, Deshmukh A (2015) Implementation and analysis of k-means and fuzzy c-means
clustering techniques. Int J Ind Electron Electr Eng 3:173–176
6. Wang YF, Wang DH, Chai TY (2013) Active control of friction self-excited vibration using
neuro-fuzzy and data mining techniques. Expert Syst Appl 40(4):975–983
7. Asish Vardhan K, Thirupathi Rao N, Naga Mallik Raj S, Sudeepthi G, Divya, Bhattacharyya
D, Kim T (2019) Health advisory system using IoT technology. Int J Recent Technol Eng
7(6):183–187. Retrieved from www.scopus.com
8. Eali SNJ, Bhattacharyya D, Nakka TR, Hong S-P (2022) A novel approach in bio-medical image
segmentation for analyzing brain cancer images with U-NET semantic segmentation and TPLD
models using SVM. Traitement Signal 39(2):419–430. https://doi.org/10.18280/ts.390203
9. Chidambaranathan S (2016) Breast cancer diagnosis based on feature extraction by hybrid of
k-means and extreme learning machine algorithms. ARPN J Eng Appl Sci 11:4581–4586
Repercussions of Incorporating Filters
in CNN Model to Boost the Diagnostic
Ability of SARS-CoV-2 Virus Using
Chest Computed Tomography Scans
Dhiren Dommeti , Siva Rama Krishna Nallapati ,P.V.V.S.Srinivas ,
and Venkata Naresh Mandhala
Abstract A pandemic called COVID-19 has threatened the world with its high
morbific and transmission rate. It is vital to accurately detect and determine the traces
of the infection as it caused around 62L deaths. Numerous researchers have set out
to propose a virus detection solution using chest computed tomography scans based
on deep learning. Yet, an accurate comparison of these techniques is not available.
Within this document, a convolutional neural network model is suggested that assists
in enhancing accuracy to detect the virus using CT scans by incorporating distinct
filters into the model. The proposed model has attained accuracy of 0.86 unfiltered,
incorporating Gabor filter helped achieve an accuracy of 0.93, and an accuracy of
0.85 is attained using bilateral, non-local means, and hybrid filtering techniques.
Keywords COVID-19 ·CNN ·Filter ·CT scans ·Diagnosis
1 Introduction
2019-nCoV is a contagious illness. It is spawned by the SARS-CoV-2 virus. The
massive public health crisis was caused by the disease and hence was announced
as a global pandemic by the World Health Organization [1]. Majority are infected
with mild-to-severe respiratory illness. By 20 May 2022, 52.4 Cr cases have been
confirmed and more than 62.7 L deaths [2]. Some get seriously ill with severe respi-
ratory symptoms which may lead to acute respiratory distress disorder, which can
lead to death. In aged patients, underlying diseases such as cardiovascular disease,
diabetes and chronic pulmonary disease, or cancers are prone to have severe health
issues; hence, mortality rate is high in these patients. The virus spreads from the
mouth of an infected individual or nasal discharge through their cough, sneeze, or
D. Dommeti (B) · S. R. K. Nallapati · P. V. V. S . Sr i n i v a s · V. N. Mandhala
Department of Computer Science Engineering, Koneru Lakshmaiah Education Foundation,
Vaddeswaram, Andhra Pradesh, India
e-mail: dhiren2910dommeti@gmail.com
P. V. V. S . S r i n i v a s
e-mail: cnu.pvvs@kluniversity.in
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_22
205
206 D. Dommeti et al.
breath, respiratory droplets to smaller aerosols. Such patients should stay at home and
self-isolate until they recover to avoid spreading the disease. CT scans also known
as chest computed tomography scans are an important approach to diagnose SARS-
CoV-2 virus. This is a powerful tool for rapid screening in suspected cases along with
other laboratory tests. Technique like RT-PCR has become a standardized detection
approach for SARS-CoV-2 as it is performed specifically based on the virus. The
limitations of RT-PCR include a sensitivity rate that is low and holds a range of 60–
70%, diversity of test methods, expenses being high, long turnaround, and limited
test capability in many countries [3]. Modern research and experimentations have
shown that clinical pulmonary images can be used to investigate for COVID-19.
Imaging modalities that are widely used for diagnosis are chest X-ray (CXR) also
commonly known as chest film and chest computed tomography scans. Due to their
wide availability and usage, the diagnostic performance of COVID-19 detection has
improved in hospitals [4]. An investigation conducted on a set of 1014 COVID-19
patients [5] states that 601/1014 patients tested positive for the virus using RT-PCR,
whereas 888/1014 patients were detected positive using computed tomography scans
(CT scans). Statistically, 59% of patients were diagnosed positive using RT-PCR, and
88% of patients were positive according to CT scan. Hence, CT scans are strongly
recommended for patients who have a negative PCR test with symptoms of the virus
[6].
2 Related Work
Evidently, there are many proposals on image analysis techniques for SARS-CoV-2
virus prognosis using ML-based algorithms to assist medical practitioners to diag-
nose and treat efficiently to detect SARS-CoV-2 virus using radiographic images of
the chest. A model called DeTrac was proposed by [7]. It is a CNN-based model that
uses the class decomposition technique to observe class boundaries and detects the
irregularities in the X-ray image. Consequently, 95.12% recognition accuracy was
observed in the proposed system. To estimate severity from chest computed tomog-
raphy images, Tang et al. [8] proposed an ML-based technique with a classification
accuracy of 87.5%. Combining methods of intelligence similar to learning from
unlabelled data commonly known as self-supervised learning and intelligence of
using previously learnt data and also a self-trans approach was proposed by He et al.
[9]. The proposed approach was to generate potent and impartial characteristics that
help to achieve 86% identification accuracy. Ucar and Korkmaz [10] proposed using
SqueezeNet for COVID-19 diagnosis. Using ImageNet dataset, the SqueezeNet was
pre-trained and fine-tuned. This was done in an augmented COVID-19 dataset, which
incorporated three types of images, COVID-19, normal, and non-infected type. For
obtaining best hyperparameters, a Bayesian optimization technique was induced in
the model which was stochastic and attempted to minimize a scalar objective func-
tion in a bounded domain. The Bayesian-SqueezeNet-based diagnosis model that was
proposed attained an accuracy of 98.26%. Narin et al. [11] proposed three 2D CNNs
Repercussions of Incorporating Filters in CNN Model … 207
for COVID-19 prognosis. Zhang et al. [12], for fast and reliable screening, introduced
a deep anomaly detection model. Ghoshal and Tucker [13] used drop weights-based
Bayesian CNN on X-ray images to investigate the estimation of uncertainty, whereas
[14] implemented both segmentation and detection using both radiographic images
and computed tomography scanned images. Fang et al. [15] reviewed the travel
records and symptoms faced by two patients, concluding that CT scans were more
accurate towards the detection of COVID-19, and the sensitivity was much higher
than the results of reverse transcription-polymerase chain reaction. Berheim et al.
[16] investigated CT scans of 121 patients’ who were tested positive for the COVID-
19 virus. Their investigation states that the seriousness of the condition gradually
increased as time went on from the initial diagnosis of the symptoms related to the
disease. Li et al. [17] put forward COVNet the deep learning model to investigate
COVID-19 by extracting visual features to differentiate between pneumonic and
other non-pneumonic lung disorders from chest tomography scans. Yet, COVNet
was unable to succeed in categorizing the seriousness of the disease. Wang et al. [18]
developed a DL-based prediction model. The accuracy obtained from the model is
89.5%. Respectfully, the approach is finer than Xu et al. [19] model which obtained
a maximum accuracy of 86.7% as it saves time for diagnosis. Xu’s model is a CNN-
based model to discriminate COVID-19 pneumonia and influenza-A. [20] trained a
U-Net++ to determine SARS-CoV-2 patients by gathering image slices of 46,096
CT imagings of COVID-19 affectees and affectees of other pneumonic lung disor-
ders. The results of their trained model accomplished good diagnosis equivalently
to expert radiologists. A distinct model was proposed with minimum layers and
a CNN architecture based on weighted filter which help result to increase accu-
racy by prioritizing a set of features [22]. CNN models are used for representation
learning as these were constrained of feature optimization. A feature selection bi-
stage approach was proposed for choosing minimum features attained from CT scan
images trained CNNs [23]. Likewise, many network architectures have also been
considering developing an AI-based detection system for COVID-19 virus.
3 Proposed Work
Through the conducted research, it is intended to classify proposed COVID-19 model
in chest computed tomography scans without any filter and state the observations
occurred when four distinct filters are applied. Framework of convolutional neural
network model being proposed consists of three fully con5nected layers: one flatten
layer and two dense layers. The first fully connected layer consists of one Conv2d
layer, the second fully connected layer consists of Conv2d layer, max pooling and
batch normalization layer each, the third fully connected layer consists of one Conv2d
layer, and the fourth fully connected layer consists of convolutional layer, batch
normalization, max polling, and dropout layers of one each. The first dense layer
consists of dense and dropout layers of one each, and the second dense layer consists
of only one dense layer. All the convolutional layers used a kernel size of 3 × 3,
208 D. Dommeti et al.
Table 1 Variables achieved by proposed model with different filters
S. no Proposed model
incorporating
Train accuracy Test accuracy Train loss Test loss
1 No filter 0.99 0.86 0.04 0.6
2Gabor filter 0.99 0.93 1.48 1.73
3Bilateral filter 0.99 0.85 0.13 0.66
4Non-local means
filter
0.99 0.85 2.09 2.43
5Hybrid filtering 10.85 1.99 2.57
max polling used is 2 × 2, and the activation functions used are ReLu in the hidden
layers and softmax in the final dense layer. The total parameters used are 32,116,743,
out of which, 32,116,103 are trainable and 640 are non-trainable. All the images are
normalized and standardized by using standardization and normalization techniques
and then resized into a fixed dimension of 48 × 48 to maintain uniformity. The dataset
contains three types of data, negative, positive, and images of uncertain category. Each
category includes approximately 5000 imagings of scanned data which are resized
to 128 × 128 pixels in the format of JPEG. The imagings are then arranged in a ratio
of train to test as 80:20. Then, filters are applied to mitigate noise. Time required per
each epoch on an average is 7 s (Table 1).
3.1 Algorithm
Input: Dataset containing images of covid19
Output: Classify the input into positive or negative or uncertain
Begin
if size(covid19_Dataset[] /= Ø) then
for all images in covid19_Dataset
JPG(covid19_Dataset[], JPG)
Covid19_Dataset[] = GaborFilter(covid19_Dataset[])
Resizing(covid19_Dataset[], 128, 128)
Endloop
Endif
Covid19_Dataset_CNN (covid19_Dataset[])
Training_Dataset, Testing_Dataset → Split(covid19_Dataset[], 80,20)
Shuffle (Training_covid19, Testing_covid19)
Covid19_Dataset_CNN → Covidl9_Dataset_CNN (Training_covid19)
Evaluation → Covid19_Dataset_CNN (Testing_covid19)
Return positive/negative/uncertain
End
Repercussions of Incorporating Filters in CNN Model … 209
4 Results and Analysis
The above proposed novel CNN model, to detect SARS-CoV-2 virus, is implemented
in five stages of modules: pre-processing, classification, mitigation, implementation,
and validation. The collected chest X-ray image from the COVID dataset [21]is
resized into pixel value of 128 × 128. Secondly, Gabor, non-local means, and bilateral
image filtering techniques are applied on the dataset. Divide the resulted datasets
in the ratio of 80:20 for training and testing. The CNN model is trained with the
training image data on all the datasets separately. The performance of the proposed
CNN model is evaluated on filtered and non-filtered images test images separately.
In Fig. 1, the proposed model shows less reduction of model loss from 60 epoch
which shows it is learning in a faster rate. The accuracy is improved with increase
in epoch, the attained accuracy is 0.86, train loss is 0.04, and the t est loss is 0.6. In
Fig. 2, model accuracy attained by the model put forward incorporated with Gabor
filter shows that as epoch increases, the accuracy also increases. The accuracy when
this filter is used is shown to be better than that of Fig. 1; yet in Fig. 2, we observe the
model loss is stabilized higher than Fig. 1, the model attained a stable accuracy 0.93
with a descent of the train loss 1.48, and the test loss is 1.73. According to Fig. 3,the
observation from model accuracy attained by the model put forward incorporating
bilateral filter states that it is like that of the model, which was unfiltered. Projecting
an accuracy 0.85 with a train loss 0.13 and test loss 0.66. Unlike Fig. 4, incorporating
non-local means filter has a similar accuracy 0.85 with train loss 2.09 and test loss
2.43 with high loss as increase in epoch. The loss is stabilized yet causing high
losses in the record. Figure 5 shows hybrid filter which is an amalgamation of Gabor
filter, bilateral, and non-local means filter techniques. The amalgamation provides
an outcome with accuracy 0.85 with a train loss 1.99 and test loss 2.57. Accuracy
is similar to filtering techniques like bilateral and non-local means yet contains high
reduction of loss in model when compared to other filter incorporated models similar
to Fig. 4. Proving the model proposed when incorporated with Gabor filter gives better
accuracy when compared to other filter techniques.
Fig. 1 Model accuracy attained by the model put forward unfiltered
210 D. Dommeti et al.
Fig. 2 Model accuracy attained by the model put forward incorporated with Gabor filter
Fig. 3 Model accuracy attained by the model put forward incorporating bilateral filter
Fig. 4 Model accuracy attained by the model put forward incorporated with non-local means filter
4.1 Dataset Execution
The datasets utilized in this study are accessed from [22]. The model being put
forward is incorporated with filters like Gabor filter, bilateral filter, non-local means
Repercussions of Incorporating Filters in CNN Model … 211
Fig. 5 Plot of validation accuracy and loss values
filter, and hybrid filtering techniques which are implemented using the Kaggle plat-
form. The execution of the model requires 13 gigabyte random access memory and 16
gigabytes of graphics processing unit and three central processing units. By applying
filters, the datasets executed are categorized as train and test divisions by 80:20 ratio,
and filters are then incorporated on training and testing modules to determine the
repercussions. Time required per each epoch is 7 s.
4.2 Graphical Representation of Accuracy and Loss Attained
See Figs. 1, 2, 3 and 4.
4.3 Comparing the Model with Latest Methods
The model is studied in detail, and the effectiveness of the suggested model is
compared with other methods. Table 2 exhibits in detail the train and test accuracies
and losses. With the dataset, the proposed model incorporated with the Gabor filter
achieves the highest accuracy of 0.99; consequently, VGG16 has attained an accu-
racy of 0.94, VGG159 has attained 0.92 accuracy, and comparatively, the accuracy
attained in these models is higher than Xception model which attained an accuracy of
0.46. From the results acquired, it can be concluded that model architecture incorpo-
rated with Gabor filter provides better accuracy comparatively. Observation depicts
that in VGG19 model as the epochs increase, the accuracy attained also increases and
model loss decreases. VGG16 model accuracy increases as epochs increase and vice
versa with model loss. The research conducted and tested concludes that Xception
model is unstable and provides a bad accuracy comparatively.
212 D. Dommeti et al.
Table 2 Performance of model with Gabor filter in comparison with various trained models
S. No. Model Train accuracy Test accuracy Train loss Test loss
1Proposed model with
Gabor filter
0.99 0.93 1.48 1.73
2VGG19 0.92 0.91 0.22 0.24
3VGG16 0.94 0.91 0.19 0.25
4Xception model 0.46 0.59 1.02 0.85
4.4 Graphical Representation of Comparing the Model
with Related Methods
See Fig. 5.
5 Conclusion
The dataset contains approximately 5000 images of three categories of data: negative,
positive, and uncertain COVID-19 CT scan imagings. Firstly, the pictures are scaled
to 128 × 128 pixels in JPEG format to be classified in a ratio of 80:20 as train and
test categories. The model suggested accommodates six convolution layers, accom-
modating six pooling layers with a dense layer. By mitigating the scale of the feature
map, the pooling layer tends to summarize the features. Hence, the feature map
is generated by the convolution layer. The filters are applied consecutively on the
datasets that are classified as trained and test modules. Finally, the suggested model
is tested with test set after being trained with train set. In the above proposed novel
CNN model, using no filter, the attained efficiency is 0.86, and train loss is 0.04 and
the test loss is 0.6 as stated in Fig. 1. Using Gabor filter, the model attained a stable
accuracy 0.93 with a descent of the train loss 1.48, and the test loss is 1.73 as stated in
Fig. 2. With bilateral filter, the accuracy being achieved is 0.85 with a train loss 0.13
and test loss 0.66 as projected in Fig. 3, with non-local means filter achieves accuracy
0.85 with train loss 2.09 and test loss 2.43 as represented in Fig. 4. Implementing
hybrid filtering, an accuracy 0.85 with a train loss 1.99 and test loss 2.57 was attained
as projected. On implementing Gabor filter, finer results are obtained comparatively,
and the detailed analogy between all models can be understood based on Tables 1
and 2.
Repercussions of Incorporating Filters in CNN Model … 213
References
1. Cucinotta D, Vanelli M (2020) WHO declares COVID-19 a pandemic. https://doi.org/10.23750/
abm.v91i1.9397
2. WHO (2021) WHO Coronavirus (COVID-19) dashboard. Available at https://covid19.who.
int/
3. Long C, Xu H, Shen Q, Zhang X, Fan B, Wang C, Zeng B, Li Z, Li X, Li H (2020) Diagnosis
of the coronavirus disease (COVID-19): rRT-PCR or CT? Eur J Radiol 126:108961. https://
doi.org/10.1016/j.ejrad.2020.108961
4. Kim H, Hong H, Yoo SH (2020) Diagnostic performance of CT and Reverse transcriptase
polymerase chain reaction for coronavirus disease 2019: a meta-analysis. Radiology 296:E145–
E155
5. Ai T, Yang Z, Hou H, Zhan C, Chen C, Lv W, Tao Q, Sun Z, Xia L (2020) Correlation of chest
CT and RT-PCR testing in coronavirus disease 2019 (COVID-19) in China: a report of 1014
cases. Radiology 296(2):E32–E40. https://doi.org/10.1148/radiol.2020200642
6. Kanne J (2020) Chest CT findings in 2019 novel coronavirus (2019-ncov) infections from
Wuhan, China: key points for the radiologist. Radiology 295(1):16–17. https://doi.org/10.1148/
radiol.2020200241
7. Abbas A, Abdelsamea MM, Gaber MM (2020) Classification of covid-19 in chest x-ray images
using detrac deep convolutional neural network. arXiv:2003.13815
8. Tang Z, Zhao W, Xie X, Zhong Z, Shi F, Liu J, Shen D (2020) Severity assessment of coronavirus
disease 2019 (COVID-19) using quantitative features from chest CT images. arXiv:2003.11988
9. He X, Yang X, Zhang S, Zhao J, Zhang Y, Xing E, Xie P (2020) Sample-efficient deep learning
forcovid-19 diagnosis based on CT scans. medRxiv
10. Ucar F, Korkmaz D (2020) COVIDiagnosis-Net: Deep Bayes-SqueezeNet based diag-
nosis of the coronavirus disease 2019 (COVID-19) from X-ray images. Med Hypotheses
140(2020):109761
11. Narin A, Kaya C, Pamuk Z (2020) Automatic detection of coronavirus disease (COVID-19)
using X-ray images and deep convolutional neural networks. http://arxiv.prg/abs/2003.10849
12. Zhang J, Xie Y, Li Y, Shen C, Xia Y (2020) COVID-19 screening on chest X-ray images using
deep learning based anomaly detection. [Online]. Available: http://arxiv.org/abs/2003.12338
13. Ghoshal B, Tucker A (2020) Estimating uncertainty and interpretability in deep learning for
(COVID-19) detection, pp 1–14. http://arxiv.org/abs/2003.10769
14. Alom MZ, Rahman MMS, Nasrin MS, Taha TM, Asari VK (2020) Covid MTNet: Covid-19
detection with multi-task deep learning approaches
15. Fang Y, Zhang H, Xu Y, Xie J, Pang P, Ji W (2020) CT manifestations of two cases of 2019
novel coronavirus (2019-nCoV) pneumonia. Radiology 295(1):208–209
16. Bernheim A, Mei X, Huang M et al (2020) Chest CT findings in coronavirus disease-19:
relationship to duration of infection. Radiology. https://doi.org/10.1148/radiol.2020200463
17. Li L et al (2020) Artificial intelligence distinguishes COVID-19 from community acquired
pneumonia on chest CT. Radiology. https://doi.org/10.1148/radiol.2020200905
18. Wang S, Kang B, Ma J, Zeng X, Xiao M, Guo J,Cai M, Yang J, Li Y, Meng X, Xu B (2020)
A deep learning algorithm using C T images to screen for corona virus disease (COVID-19),
1–26. medRxiv preprint. https://doi.org/10.1101/2020.02.14.20023028
19. Xu X, JiangX,MaC,DuP,LiX,LvS,YuL,ChenY,SuJ,LangG,LiY,ZhaoH,XuK,Ruan
L, Wu W (2020) Deep learning system to screen coronavirus disease 2019 pneumonia, 1–29
20. Zhou Z, Siddiquee MMR, Tajbakhsh N, Liang J (2019) UNet++: a nested U-Net architecture
for medical image segmentation. IEEE Trans Med Imaging 3–11
21. https://www.kaggle.com/datasets/andrewmvd/covid19-ct-scans
22. Sanghavi F, Panetta K, Agaian S (2021) Covid-19 detection in CT images using custom
weighted filter-based CNN. In: Multimodal image exploitation and learning 2021. International
Society for Optics and Photonics, p 117340L
23. Sen S, Saha S, Chatterjee S, Mirjalili S, Sarkar R (2021) A bi-stage feature selection approach
for Covid-19 prediction using chest CT images. Appl Intell 1–16
Software Development Estimation Cost
Using ANN
Puppala Ramya , M. Sai Mokshith, M. Abdul Rahman, and N. Nithin Sai
Abstract There are several inter-related aspects that have an impact on the effort
and productivity of software development. Since the majority of these connections
are not well understood and are impossible to predict accurately, software develop-
ment time and effort have always been a challenging undertaking. In use or suggested
in the literature, regression-based estimating models predominate. The study looks
into the potential of software artificial intelligence methods for creating the following
software development effort estimation models: case-based reasoning with artificial
neural networks. When there are intricate relationships between variables, artificial
neural networks are capable of providing accurate estimation. Numerous intercon-
nected aspects that are involved in software development have an impact on both
the development- effort and its productivity. Because more of the relationships were
not healthy. The research examines the potential of these 2 artificial intelligence
approaches, that is, artificial neural networks (ANNs) and case-based reasoning
(CBR) to creating development effort of estimation model.
Keywords Artificial neural networks ·Case-based reasoning ·Software
development
1 Introduction
Software development estimation is a complex problem that has attracted a lot of
research interest in trying to make the techniques project managers have access to for
estimating information more useful in terms of time management. Software develop-
ment estimation involves the number of these inter-related factors that affects devel-
opment effort and its productivity, and to accurate forecasting have been challenging
because all these relationships have not been well understood.
P. Ramya (B) · M. Sai Mokshith · M. Abdul Rahman · N. Nithin Sai
Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation,
Vaddeswaram, India
e-mail: mothy274@kluniversity.in
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_23
215
216 P. Ramya et al.
Fig. 1 Software estimation cost
Regression approaches are used in the major majority of the estimation models
which are in use, or which are proposed in the literature, this research investigates the
utility of 2 artificial intelligence methodologies, namely artificial neural networks and
case-based reasoning in estimating development effort estimate estimation models.
Artificial neural networks are well known acknowledges the ability of delivering
very effective situations containing complex interactions between inputs and outputs
as well as input data affected by a lot of noise these features establish the soft-
ware development environment where development estimates are calculated although
neural networks have a tremendous potential for accuracy prediction delk explana-
tion capabilities and do not provide direct user adoption of outcomes one or more
cases are discovered by CBR that they are compatible to current problem and try to
change them to meet the parameters of the current situation each case in the soft-
ware development work estimates could be a prior software development, however,
the present difficulties in determining an appropriate estimate to the current project
case-based reasoning can be used to justify judgments based on past incidents that
have been utilized to solve an issue (Fig. 1).
2 Estimation Model Performance
Variety of software-effort estimation model have been studied by a number of
researchers, and the problem has been revealed to be a complex one with unfa-
vorable outcomes in general. Kemmerer used data front projects outside of the orig-
inal model development contexto empirically validates 4 algorithmic model (SLIM,
COCOMO, estimates, and FPA), Without Salepur rating the models, the findings
show how general these models are across different situations, the mean for this
absolute relative error ranged from 57% to over 80% in most models indicating a
substantial overestimation bias and large estimate mistakes (Fig. 2).
Dr. Ferens and Gruner used all the projects from the Albrecht’s, database at 14:00
from Kemerer datasets shows test development and effort prediction model such as
SPANS, checkpoint, and COATAR. The MARE ranges from 46% from the check-
point model to 185% for the coaster model indicating high prediction error a study by
Jeffrey lo looked into the requirement for model calibration at both the industrial and
organizational levels even again the mail is so high once again which is raining from
the number 43 to one or 5% for 3 companies studied using get up from 64 projects
Software Development Estimation Cost Using ANN 217
Fig. 2 Estimation model
performance
within one organization Jeffrey, Low, the Barnes compared to the SPQR/20 models
to FPA 2 eliminate over are underestimation BIOS is the models were recalibrated to
the local environment the estimation errors or for lower than in earlier investigations
with maids of around 12% demonstrating the benefits of model calibration MSRA
studied 364 forms and discovered that just 51 utilized model 2 estimation of software
-development effort and those model users estimates work know better than the no
model users estimations Heemstra discovered that most generic models are utilized
without recalibration and that most models are based on these generic models.
3 Artificial Neural Network Models
Have the capacity to comprehend complex relationships that are able to measure the
function, which suggests that application success is due to sudden learning insuffi-
cient numbers of hidden units are lacking with the fixed relation between IP and target
has many features that make it interactive. Pattern recognition is based on tasks with
building eligible system rushes the network itself is a model because the topology
and node distribution network we used to measure the software development effort
in this study (Fig. 3).
• Software development involves in great number of inter-related factor that affect
development efforts and productivity due to many of those Working relation-
ships are not unlisted. The topic of accurate software development time and effort
prediction is challenging 3 attributes: Every example also includes mean and
median values, with mean values greater than median values indicating that the
sample numbers are biased toward lower or smaller production enterprises. When
dealing with issues where there is a complex link between important output and
where the input data is corrupted by high noise levels, artificial neural networks
(ANNs) are known for their capacity to produce good results. Despite the fact
that there is good potential for predictive accuracy, the property characterizes
218 P. Ramya et al.
Fig. 3 Artificial neural network
the software development environment from which the development effort esti-
mates, or generation is derived networks of neurons lack of direct user adoption
environments and limited explanatory skills.
4 Network Performance Evaluation
4.1 Performance Measurement
The vivid errors and estimates we use are the different people pulse which have been
went through the researchers metrics, but the project will be the main measure of
model performance most related error, it is the preferred way of measuring errors by
the software researchers hello add ditis calculated by the following (Fig. 4).
Where the network output for these each of the observation is estimation.
Software Development Estimation Cost Using ANN 219
Fig. 4 Performance evaluation
The quantity of observations is N.
The major relative error is assessed to determine if the model is biased toward
overestimating or underestimating the weather.
4.2 Simulated Development Data
Talk Balaj all the development effort estimation ability of a nice huge massive
complex environment for the train to set with the set was required to allow which
will make use of large enough to permit all these n/w in order to the capturing of
all problem dominate in the minimum of the number required one in observation in
order to achieve satisfactory genre: off the fraction for the error in the trend set which
will be less than all. The errors fraction is to be assumed that is smaller than 0.1254.
This will play implies that all the guidelines approximately 10 observations or to
be required with all of each rates in order to connect the network. In these sections,
all neural networks will be gained through an environment where it includes much
development attributes, this will require a quite massive development of software
small fulfills the entire recruitment discussed in above. These suitably a great dataset
which will be for the been restricted to some development project which are having
some liable door code the entire enormous development attributes which will have
further been included in model. And alternate approaches employed to enable such a
study in which simulator software development project data is generated, and this is
done with the SPQR/20 software estimation tool. The output of SFQR/20 has been
proven to be reasonably accurate in estimating development effort in locally cali-
brated context meaning that the output in produces is indicative of the development
220 P. Ramya et al.
Fig. 5 Shows histogram of
average errors
15
10
5
0
20
25
30
0.02 0.06 0.1
ARE
Histogram
environment, and all the input values for the simulated project data were generated
by using the random number-generator.
Data for 1000 businesses was created and manually entered into SPQW/20 to
estimate project development. Wide variation of system size (109–15,571 FPS), 8
development hours, and little value (2162–912,309). Although this size range was
produced, it was difficult to scale because neural networks’ predictions, skill input,
and output networks. It is quite interesting to see the productivity range that SPQW/20
generate. Projects completed with this high productivity are approximately 10 times
more productive than projects completed with the lowest productivity. For the neural
network to calculate an effort to estimate from these function point-size for the main
development attribute impacts, the development productivity range indicated here
must be reflective of commercial development. Figure 5 displays a histogram of
average errors.
Figure 5 shows a 42 histogram of average relative errors to highlight network
performance because the test set has 100 observations, the frequency on the WI axis
also shows frequency percent, the frequency of estimates returns of less than or equal
to 2% is indicated in first column, the frequency of equals 4, 6, 8 and 10% is noted in
the other columns, and the Victorian department of the Australian-software-metric
association has created (ARE) to show network performance (Fig. 6).
5 Conclusion
In a huge dataset of these simulated projects data with significantly less than generally
happens in our project data, ANNs were successful in properly forecasting project
work. This dataset has enough observations to allow for the adequate training. The
obvious motion is that when the datasets noise level in rises, estimation mistakes
will rise as well. Back propagation network for able to predict development effort of
over 25% of actual effort than that of 75% of the project in the test sets in ASMA
Software Development Estimation Cost Using ANN 221
Fig. 6 ARR to show network performance
dataset, With the mayor of less than 0.25, even still as noted by Baum and Haussler
[2] and Hinton [5], this dataset does not completely matched the data-needs of neural
networks. The usage of car reader less favorable results within 11 of 15 examples
(73.3%) falling within 50% of actual effort value and 8 of 15 (53.3%) falling with
25%. Optimistic, however, because only a few parameters were taken into account
for adaptation, and the criteria applied were relatively narrow win scope artificial
neural network shaved Mann stated their capacity to generate an effective effort
estimation model despite the limitations of the provide dataset. Although CBR looks
to have promise, more research is needed to improve the estimation model quantity.
If the current rate of project growth within the ASMA dataset continues, the database
should offer foundation for the creation of new estimation models within a few years.
Bibliography
1. Kocaguneli E, Tosun A, Bener A (2010) AI-based models for software effort estimation. In: 36th
EUROMICRO conference on software engineering and advanced applications, pp 323–326.
https://doi.org/10.1109/SEAA.2010.19
2. Verma S, Popli R, Kumar H, Tanwar R (2021) Futuristic trends in network and communication
technologies, vol 1395, p 556
3. Finnie GR, Witttig GE (1996) AI tools for Software Development and Testimation. In: Proceed-
ings of the 1996 International conference on software engineering: education and practice, pp
346–353. https://doi.org/10.1109/SEEP.1996.534020
4. Estimating function points: using machine learning and regression models. In: 2nd International
conference on education technology and computer (ICETC) 2010, vol 3, pp. V3-52–V3-56
(2010)
5. Kaushik A, Soni AK, Soni R (2012) An adaptive learning method to software cost estimation.
In: National conference on computing and communication systems (NCCCS) 2012, pp 1–6
222 P. Ramya et al.
6. Khan MJ Applications of case-based reasoning in software engineering: a systematic mapping
study. Software IET 8(6):258–268
7. Popli R, Chauhan N Costing effort estimation in Agile software development. In: Optimisation
reliability and information technology
8. Bagga G, Bajwa MS, Rana R (2021) Recent Innov Comput 701:13
9. Pinzger M, Macho C, Beyer S, McIntosh S (2021) The nature of build changes. Empirical Softw
Eng 26
A Generic Flow of Cyber-Physical
systems—A Comprehensive Survey
Jampani Satish Babu , Gonuguntla Krishna Mohan , and N. Praveena
Abstract The overwhelming trend of modern technology is studious as the conven-
tional systems and mobile devices evolve as intelligent devices and intelligent
systems. Cyber-physical systems (CPSs) have emerged as the crucial factor in the
real-world environment with various essential requirements. Wireless systems and
IoT-based computations are considered to attain a similar interaction among the
computer networks and humans, opening up diverse opportunities and challenges for
modeling efficient cyber-physical systems (CPS). It considers human factors during
the process of system operation and real-time management. This work provides a
comprehensive survey of specific current research that facilitates CPS. The crucial
factors like development, applications, architectural design, real-time cases, stan-
dards, provisioning techniques, and network for CPS are discussed. Here, integrated
framework and virtualization approaches of networking, computing, caching are
examined to offer a baseline the innovative world development. The performances
of specific models are discussed with their pros and cons to enhance the model.
Finally, some research issues are discussed, and possible outcomes are discussed to
help the young researchers to make further research flow.
Keywords Cyber-physical systems ·Internet of things ·Wireless system ·
Intelligent devices ·Virtualization model
1 Introduction
The drastic innovations over the technological advancements have the people’s
lifestyle in agriculture, industrial application, medical treatment and transportation,
and other essential things [1]. Similarly, some newer trails are also launched by
J. S. Babu (B) · G. K. Mohan
Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation,
Vaddeswaram, Guntur, A.P 522302, India
e-mail: jampanisatihsbabu@gmail.com
N. Praveena
VR Siddharth Engineering College, Kanuru, Vijayawada, AP, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_24
223
224 J. S. Babu et al.
replacing conventional physical devices. By considering an information system as a
productive example, it is easier to predict the traditional system of information that
generally relies on the embedded devices with near features [2]. These systems lack in
fulfilling the essential application requirements of the physical devices-based control
and interaction [3]. Thus, with the advancements in modern information technology,
CPS slowly turns into a mainstream process for technological progress to substitute
the conventional information system [4].
Generally, CPS is an intelligent system composed of the controller, actuators, and
sensors embedded over another method to assist the interaction among the cyber
world and the physical world. It is composed of three diverse layers like perception
layer, transport layer, and application layers [5]. The first layer is utilized to attain
the required sensitive information and provide superior feedback decisions. The next
layer includes both the physical and network layers that act as a medium access
control mechanism adopted to transfer the essential information and make better
decisions over diverse system elements [6]. The last layer is observed as the control
layer, which is wholly adopted for making a better decision based on the broader
results attained from the information perspective. This model with the layered view
is extensively adopted in diverse CPS [7]. These are explored in various real-time
applications like smart grids, transportation, industries, electric vehicles, etc.
When the network applications over the conventional embedded information
system pretend to act more comprehensively, the investigators emphasize uniting
the human requirements and networking model. Therefore, it realizes the interac-
tion between the network and the physical society [8]. In the provided context,
cyber-physical social systems have emerged in recent days, which are also termed
as the association system that merges the human resources, physical, and computa-
tional resources. Also, it facilitates the coordination between the social, physical, and
cyber worlds [9]. It assists in parallel execution, self-synchronization, and supervi-
sory measure over the social domains, cognitive measures, material and information
domains [10]. Therefore, it is competent to offer an intellectual paradigm to attain
the constructive and the design goal of the intelligent environment with control and
command organizations.
Contrary to the CPS, CPSS is considered humans as the system part and merges the
human into the loop condition. CPS considers some human factors via the intellec-
tual human–computer interactions during the system-level functionalities [11]. Thus,
these factors are depicted as the superior supervision process of modern systems.
Humans-CPS models are determined or viewed as both the service providers and
the service consumers, unlike the management and operations of the conventional
CPS. It has to examine the individual’s competency before determining the humans
over the loop of execution over the given context [12]. It is essential to recognize
the perspective of an individual to choose and carry out the task performance. The
mainstream confronts while merging the social space from the human activities is
divergent from the computers’ perspective. Individuals cannot notice this sort of
functionality all the time, and sometimes, they do not select the appropriate instruc-
tions without proper notifications [13]. In contrast to the computerized models, it is
A Generic Flow of Cyber-Physical systems—A Comprehensive Survey 225
Fig. 1 Generic view of cyber-physical systems
known that the individuals are more reliable; moreover, they have the better compe-
tency to adapt to various environmental changes dynamically and intend to offer
multiple innovative outcomes [14]. Thus, the developers need to consider the human
characteristics during the optimization and construction of CPS [15]. Therefore, it
facilitates the interaction among the devices, computers, and individuals (users).
Figure 1 depicts the generic view of CPS.
This work attempts to offer a comprehensive survey of the various prevailing
models related to CPS with this extensive analysis. Even with the further exten-
sive comments and reviews over the intelligent systems, the investigators are inter-
ested in the study of CPS [16]. Some reviewers discuss CPS at the application
level over specific fields like industrial Internet, platoon-based vehicular models,
wind energy conservation systems, smart grids, healthcare systems, and conversion
systems with various prevailing researches. Some researchers concentrate on specific
research-based issues and the approaches adopted by diverse CPS applications like
machine intelligence, security, and so on, thereby reviewing the smart cities and
intelligent healthcare systems in IoT indeed of CPS. Moreover, the former models
show constraint over the application level while the latter examines the data-centric
factors. Koscher et al. [17] discuss the recent advancements in smart grids, social
manufacturing, and intelligent transportation. These sorts of reviews offer a more
comprehensive investigation based on CPS in the specific application field. However,
the author targets the adoption of reinforcement and deep learning approaches and
emphasis that the traffic data management system is also an essential factor for
discussion.
With the significant variations with prevailing approaches, this article intends
to discuss the integration of CPS with social communication and elaborate the
constraints over specific applications. Additionally, this work concentrates on the
human functionality over the CPS; however, the discussion is not so constructive
toward the social aspects. Figure 1 depicts five diverse factors of CPS that concentrate
226 J. S. Babu et al.
on the development, applications-based case study, architectural design, facilitation
of various modern techniques, and networking model with research challenges.
2 CPS Evolution
The recent era is the CPS 2.0 era, differentiated by incorporating big data, cloud
computing, communication, computation, control, and artificial intelligence (AI)
[18]. It is referred to as the CPS 1.5 or CPS 2.0 with the name of the human-CPS
mode. Some of the characteristics of this model are listed below:
• The cyber world is constructed with the system model with control, computation,
and communication networks.
• Physical world includes electrical, mechanical, and chemical process.
• The physical side of the cyber systems is measured with actuators and sensors.
• Database servers are essential to collect and preserve the event generated by the
sensors for computational control.
• Networks connected with the cyber systems like actuators and sensors are
provided for communication establishment. The data captured by the sensors
for physical processing is transferred to the CPS and database for processing
and storage purposes. The control decisions are transferred to the actuators for
performing the control actions of the commands.
The widespread research and scaling toward technological growth provide various
application-level support, sensors, software systems, actuators, etc. [18]. Thus, CPS is
integrated with the communication, computation, and control systems. The velocity,
variety, and volume of the model are generated with PCS with all the essential
features. The conventional database servers are utilized for processing, storing, and
data usage [19]. However, it is not efficiently fulfilled by CPS 1.0. CC and big data
model assist it, and it is considered the improved CPS and gets the name as CPS 2.0.
• Big data functionality is connected with CPS as the data is collected and utilized
for modeling an algorithm. It examines the interactions and nature of the cyber
systems to eliminate and identify the abnormalities.
• Big data is used to gather environmental information and examines the system
behavior to model the self-reconfiguring components for self-adaption.
• It is used to model an intelligent control and service-based higher-level model for
the generation of the data.
• Service-based computing nature facilitates the modeling of the application based
on CPS like innovative healthcare systems, smart homes, intelligent transportation
systems, etc.
• Cloud computing helps provide the management and control of various data
services, interface devices, software services, and physical processes like infras-
tructures and ICT hardware.
A Generic Flow of Cyber-Physical systems—A Comprehensive Survey 227
Therefore, CC and big data over CPS 2.0 are determined as the critical component
to fulfill the service requirements toward the customer, system, safety, fault tolerance,
adaptability, stability, and other QoS constraints [20]. Moreover, there are diverse,
challenging factors related to the model’s reliability, security, and safety with the
increased complexity and scalability as the learning system does not validate the
system process.
3 Applications
There is no proper authoritative and general definition for CPS where it is specified
that CPS is a system used for controlling and monitoring the physical world envi-
ronment. It is determined as the newer generation of embedded control systems like
CPS-based network embedded systems. Additionally, some plans include the actu-
ators and sensor-based network model over the embedded system, which the CPS
also considers [21]. Based on the reliance on the system model, CPS is depicted as
the IT system that merges the physical world activities over the application model. It
is the outcome of the information advancements and communication technologies to
facilitate the interaction among the biological processes [22]. These sorts of defini-
tions project are the occurrence of the interactions among the physical and the cyber
world problems.
The higher dependencies over the CPS are ascending in a day-to-day manner
in diverse applications like transportation, energy, healthcare, military, and manu-
facturing industries. This CPS is provided with various names based on the appli-
cations they are adopted [23]. For instance, a most essential and significant CPS
model shows supervisory control over the data acquisition model adopted in CI
like ICS and intelligent grids. In some cases, CPS emerges as medical devices like
implantable medical devices and wearable devices. Additionally, the networking
model of the small control systems is embedded over the smart cars to enhance
safety, fuel efficiency, and convenience. This article initiates a brief representation
of the CPS application that is investigated over the section given below.
(a) Smart grids. It is considered the advancements of the next-generation power
grid system utilized for distribution, transmission, and electricity generation f or
the past few decades. It offers diverse advancements in application-level func-
tionalities and benefits and improves global load balancing, emission control,
energy savings, and intelligent generation [24]. It facilitates the home consumers
with superior control over the energy, which is environmentally and econom-
ically beneficial at the local level. It is composed of two diverse compo-
nents known as supportive infrastructure and power applications. The former
model is depicted as the intelligent component that significantly concentrates
on monitoring and controlling the core smart grid operations with the hard-
ware, software, and communication networks. At the same time, the later model
228 J. S. Babu et al.
determines the core functionality of the provided smart grids like electricity
distribution, transmission, and generation.
(b) Medical devices. The medical devices are enhanced by integrating physical and
cyber functionalities to provide superior healthcare services. More researchers
concentrate on the analysis with the medical devices with specific cyber capabil-
ities that offer physical impact over the patients. These devices are either wearied
by the patients or implanted in patients’ bodies, termed wearable devices [25].
Generally, it is equipped with wireless capabilities to facilitate communica-
tion among the devices, i.e., programmers intend to reconfigure or update the
devices. Similarly, the wearable devices are connected with the PCs over the
control center for controlling and monitoring the functionalities from the remote
location. Table 1 depicts the comparison of CPS attacks over medical devices.
(c) Industrial control systems. It specifies the control system utilized to improve
production by monitoring and controlling t he system in various industries like
sewage systems, nuclear plants, and water and irrigation systems. It is also
known as distributed control systems or SCADA. For more reliability, ICS
Ta bl e 1 Cyber-physical attacks over medical devices
Name Impact Approaches Pre-condition
DoS Individuals do not
receive any sort of
expected therapy
Re-transmit ‘turn-off ’
commands
Captures ‘turn-off’
condition transferred
by the programmers
Unauthorized
commands injections
Wrong decisions Transmitting packets
with false data and
impersonate CGM
Injecting
communication
among the pumps
False data injection Crucial health
conditions
Impersonate remote
control by transferring
packets with
unauthorized
commands
Injecting
communication
among the remote
and the pumps
Malware injection Frozen IPS and
failures over the
ON/OFF state
Communication among
the BCM
Physical access
toward the bus system
Packet injection False injection rate,
control loss, DoS, and
safety measures
Compromised packets Malware injections
Replay attack Safety control
measures
Re-transmission of
legitimate commands
and eavesdropping
Access toward the
network bus
Car spying Unauthorized access
and theft
Captures the relay
nodes and beacon
signals from car to key
and relay outcomes
signal from critical to
the car
Attack requires relay
tools like amplifiers
and antennas
A Generic Flow of Cyber-Physical systems—A Comprehensive Survey 229
discusses here with diverse controllers based on collaboration to attain various
expected goals. Some popular controllers are known for their programmable
logic controller, a microprocessor design to constantly function over the hostile
environment. It is connected with the physical world problems like actuators and
sensors [25]. Generally, it is equipped with wired and wireless communication
ability configured based on the surrounding region. Also, it is connected with
various PC systems over the control center for monitoring and controlling the
functionalities.
(d) Smart cars. It is also known as intelligent cases that are more user and
environment-friendly with safe, fuel-efficient, improved entertainment, and
valuable features [26]. These advancements are made probable with the effi-
ciency range of 50–80 networked computers put together, and it is known as
electronic control units (ECU). It is accountable for controlling and moni-
toring diverse functionalities like brake control, emission control of engines,
multimedia and radio players-based entertainments, and comfort regions with
windows operation and cruise control mechanisms.
4 CPS Models
It is a component that has the competency to communicate with other CPS compo-
nents and control centers. These components are composed of actuators, sensors
and connected with the physical world. Each of these possesses various security
measures that affect the interactions of multiple elements and their corresponding
capabilities [27]. For instance, the CPS components may communicate with certain
computational functionalities that are not expected to influence the physical world. It
is exploited to provide unexpected characteristics with physical outcomes. However,
the components’ physical properties and the objects interested in these objects in the
physical world can monitor and control the random attacks and effects in non-physical
outcomes like misleading information data transferred over the network.
The CPS components are competent to communicate with various control centers
or other elements. These components are composed of actuators or sensors that need
to be connected with the physical world components. These capabilities possess
diverse security implications that outcome in the interaction of features and capabil-
ities [28]. For instance, the components may communicate with the computational
functionalities that are not intended to influence the physical world. It is exploited by
the unexpected characteristics of the physical consequences. The components’ phys-
ical properties and the objects’ physical properties over the physical work can control
and examine the malicious attacks that affect non-physical effects like misleading
data over the network. CPS heterogeneity between the available components or over
the elements’ understanding lacks the functional features and outputs in newer kinds
of security threats that exploit the model heterogeneity. It has to differentiate various
aspects straightforwardly with security analysis. Therefore, this article proposes three
diverse elements like cyber, physical, and cyber-physical. The physical factors are
230 J. S. Babu et al.
composed of components that can interact directly toward the physical world like
actuators and sensors [29]. These properties may possess safety and security-based
issues. The cyber-physical and cyber factors are composed of anything random that
does not interact directly with the outside or physical world, i.e., communication
process, computations, and monitoring activities. These two factors possess similar
features; however, the critical variation relies on the interaction with the physical
components.
In the CPS model, the cyber component does not directly interact with the phys-
ical components; however, the cyber-physical components directly interact. These
differences assist in offering CPS security analysis over the diverse factors, and
cyber-physical aspects can connect directly with the physical and cyber world. The
industrial control system controls and monitors the temperature over the chemical
plant, where the objective is to maintain the temperature with the provided range.
When the temperature seems to exceed the given threshold level, then the PLC
provides notification through the sensors presented over the tank and notifies the
control center when the temperature varies. The PLC handles the cooling system to
diminish the tank’s climate over the desired range. The cyber interaction with the
PLC provides no direct interaction among the physical components like tank and
cooling fans. It includes direct connection with laptops and communication with
higher-end environments like r emote entities and control centers. The wireless inter-
face is based on short and long-range frequencies [30]. In smart grids, the smart
meters are attached to the utilities, measure the appropriate electricity consumption,
and monitor usage information. It is an interface between the energy management
system and the house appliances. Here, wireless is the standard means of commu-
nication. The meter is equipped with a short-range interface using diagnostic tools
and digital meter readers. The collector can transfer the aggregated data over the
designated neighborhood and manage the utility companies. Specifically, the data is
transmitted to the serves, maintains data, and shares among the management system.
It is determined as an efficient way to establish control and exploit the launch of
blackouts over the number of smart meters.
In medical devices, the implanted defibrillator is manually and automatically
injects the insulin for the diabetes patients. The ICS is adopted to predict the heartbeat
and response to preserve the heart rate. It is an insulin pump known as a constant
glucose monitor to measure the blood sugar level. The glucose level is received from
the continuous glucose monitor (CGM) and transfers through the wireless signals to
other devices. The cyber embedded systems are directly connected with the system
over the hospital to provide wireless communication. It can interact directly with
the implanted devices and specify the cyber aspects over the medical devices. In
smart cares, communication is expected through the ECU, which is connected to
the sub-network model. It possesses various sub-networks for establishing inter-
communication among the gateways. It concentrates on providing security issues
and CANs deployed over the network.
A Generic Flow of Cyber-Physical systems—A Comprehensive Survey 231
5 Security Models
With a broader relationship establishment among the computational and physical
infrastructure, the system needs to provide essential security measures. The diversi-
fication and complexity among the physical and cyber components cause the system
to be more vulnerable by provisioning enormous threats. The interruption over the
physical infrastructure causes the bad weather condition or crisis in war condition
to dispute human lives. Some attacks target the communication environment and
influence the system by maliciously affecting the approach to capture sensitive infor-
mation. These sorts of attacks are harmful and intend to control the IT system with
communication disruption and terminate the system activity. These attack types can
influence the individual over the physical environment and tamper the interaction
among the system [31]. Data analysis and system monitoring are essential for the
evaluation of threats and performance. The system can analyze the utility and moni-
toring of the cyber to observe the issues over the privacy violation process and
proprietary model. For instance, consider the power distribution problem over the
intelligent grids and evaluate the power consumption analysis over the industrial and
residential areas. It includes the classification of household appliances, time usage,
and energy consumption [31]. It does not influence the infrastructure functionality
like information with valuable content with specific property and predicts empty and
vulnerable property theft timing.
Another instance of CPS security relies on the unmanned vehicles that use the
camera that records the region to construct the terrains with appropriate functionality.
The footage captures determined as the surveillance materials that are leaked and
exposes the footage [32]. In the case of implantable medical devices, the informa-
tion gathered by the systems is composed of diagnosis measures, model identifiers,
therapy regimens, and so on. This information is not preserved more safely. The
attackers use it to target individuals. Thus, the significant objective is to offer a privacy
measure toward CPS’s security policies, and those systems need to provide reliable
and safe functionality to achieve security. Table 2 depicts the CPS vulnerabilities.
(a) Attacks on actuators and sensors. Papadimi et al. [33] discuss the mitiga-
tion problem and the intrusion detection process over the control system based
on SI, AE, and SE attacks. The attackers influence the sensors, rub the sensor
observations, and inject false statements. It leads the system to move inside an
unsafe environment. The finite-state automata for the class attacks are detected,
and the defending mechanisms are provided for the online attacks and measure
the control events after the detection process. An algorithmic model is required
to validate the system environment and protect them from the attacks where
the damages are modeled based on pre-defined accessibility of unsafe system
state. The necessary and sufficient condition is needed to fulfill the system
requirements and eliminate the damage caused due to the attacks mentioned
above. Papadimi et al. [33] discuss the defending mechanism that eliminates
the network attacks over the actuators and sensors. When the system is not
232 J. S. Babu et al.
Ta bl e 2 CPS vulnerabilities S. No. Cyber-physical systems Vulnerabilities
1Industrial control systems 1.Web-based attacks
2.Wireless and wired
communications
3.Insecure protocols
4.Insecure access points
5.Equipped physical
storage
6.Interconnection field
7.Open communication
protocols
8.Software measures
2Smart grids 1.Communication
protocols
2.Software
3.Customer’s privacy
invasion
4.Interconnected field
devices
5.Insecure smart meters
and protocols
6.Blackouts
7.Physical sabotage
3 Wearable devices 1.DoS
2.Software
3.Noise and jamming
4.Injection and replay
attacks
5.Privacy invasion
4 Smart cars 1.GPS traceability
2.Replay attacks
3.Easier interception
4.Communication flaws
in software
5.Player exploitations
6.Unprotected
components
7.Authentication flaws
8.Insecure bus system
identified with attacks, then the defending mechanism does not vary the char-
acteristics of the closed-loop system. The author introduces the detectable and
undetectable network attacks to validate the properties and provide the necessary
and satisfactory conditions for predicting the attacks. It is essential to ensure
the countermeasure of the control system and provide specific requirements for
detectable and undetectable security systems. Figure 2 depicts the CPS attack
surface.
A Generic Flow of Cyber-Physical systems—A Comprehensive Survey 233
Fig. 2 CPS attack surface
Kim et al. [34] discuss CPS under attack as a descriptor system with certain
constraints over the unknown inputs. It influences the measurement state and estab-
lishes the model for attack detection and recognizing the consequences of the attack
over the output measurements. The constraints over the class labels are monitored
with graph theory and system theory. The significant performance of the system
is to examine the defect over the physical attacks and monitor the signal that trig-
gers the system dynamics. Daneiz et al. [35] discuss modeling a powerful attack
with uncertain CPS devoid of any attacks. The zero-dynamics attacks are provided
with standard representation. It is adopted over uncertain systems, and alternative
methods are offered for providing a perfect system. The robust attack model requires
nominal plat as input and examines the output signals. As the statement provided
by the Daneiz et al. [35], a man-in-the-middle attack is such a crucial attack over
the CPS model. The intruders’ senses create, hide, or vary the information over
the sensor and manage the communication channel. Daneiz et al. [35] construct a
deterministic model over the sensor and actuator channel to defense mechanisms to
protect the system from damage. The safety control measure is determined over the
network attacks. It is termed a safe-controllability measure, predicts the attack over
the network, and stops the system from reaching the unsafe environment. This model
is provided to validate the attributes, and some computational devices are modeled
to predict the dangerous environment known as the intrusion detection model. Some
mathematical models like Bernoulli, queuing, and the Markov model are used for
examining the CPS performance when the system encounters DoS attacks.
Gupta et al. [36] discuss the risk-sensitive problem on DoS attack under Markov
modeling. The attacker needs to use the Markov model to measure the system-based
control packets. Gupta et al. [36] discuss the probability measure to examine the
234 J. S. Babu et al.
stochastic properties, and the hidden Markov model is provided to evaluate the risk-
sensitive control factors. The author discusses the consequences of DoS attacks.
The performance is measured through the linear-quadratic model to reduce the
cost function of the system over the attack environment and offers a novel solu-
tion with sensitive programming. Gupta et al. [36] discuss the DoS attack model
with certain constraints by restricting the frequency of the DoS attack. It is probable
to capture diverse DoS attacks that include random, periodic, trivial, and protocol-
based jamming attacks. The robust control measure over the DoS attack is examined
and intends to reduce the attack frequencies without any undamaging stability. The
author discusses a dynamical event-triggered control mechanism to override the
DoS attacks. When the advanced controllers intend to exchange information, the
attack measure examines the information transmission and predicts the vulnerabili-
ties. When the vulnerability is identified, the system is intruded on by DoS attacks and
moves to an unsafe environment. The fault ridding mechanism needs to be designed,
and the significance of the model should be analyzed.
6 CPS Properties
In general, CPS functions in a diverse environment to accomplish various purposes.
The functionality needs to fulfill both the physical and cyber securities. There are
three non-functional requirements related to higher-level perspectives, i.e., avail-
ability, security, and safety. These three properties are achieved with complex multi-
disciplinary systems with objective challenges and need holistic consideration. CPS
possesses diverging nature due to integrating physical and cyber components and
establishing the dynamical interaction among the physical and CPS environment
[37]. The constant variations provide aggressive variation over these three proper-
ties and establish a better trade-off among the devices. The researchers aim to offer
security with related properties. These properties are listed below:
(1) Safety. CPS operation is based on the interaction with the dynamic physical
environment and influences humans’ lives. The essential requirements of the
CPS are to fulfill the individual’s safety with the specific operation. CPS provides
intelligent context to make a better decision when physical constituents influence
the systems. Concerning safety measures, the CPS functionality is based on the
probable scenario and provides better output decisions that harm human lives.
For instance, in smart grids, the CPS should check the power and ensure no
voltage fluctuations. However, in unmanned vehicles, the malfunctioning needs
to be measured, and when the system encounters safety violations, it has to be
resolved with proper safety policies.
(2) Security. Both cyber and physical security need to be considered where the
protection toward the system components needs to be identified for theft, unau-
thorized tampering, weather condition, information integrity, and confidentiality
A Generic Flow of Cyber-Physical systems—A Comprehensive Survey 235
of the information access needs to be considered. The equipment of the intelli-
gent power grids needs to be located out and installed with weatherproofing. The
protocols and policies are determined with operational functionalities and infor-
mation monitoring and accessed in an authorized manner [37]. The failure due to
some catastrophic outcomes needs to adopt critical situations. The attacks over
the insulin pump need to identify the malicious activities and the control captured
by the adversaries and cause the death of the individuals. Sensitive informa-
tion confidentiality is associated with the physical components that need to be
preserved. The information available toward the malicious entities is exploited
with discrimination, blackmailing, theft, identity, etc. In some cases, unautho-
rized persons have the information acquisition, and the sensitive information is
stolen. The primary security feature of CPS is the cyberattacks with physical
consequences. The security model deals with cyber threats and considers both
the critical physical and cyber system processes.
(3) Availability. CPS is generally utilized for offering huge crucial functionalities
where the operation is uninterruptedly influenced for a longer time. However,
there should be a proper balance among the energy/power needed for the compu-
tational process and consumes more energy [37]. The downtimes of the crit-
ical systems are not accountable, and the vulnerabilities process is eliminated
probably. The patches trigger the functionality and tamper the operations. Due
to the available system requirements, the risk factors over the system remain
un-patched.
In CPS, the security trade-off is established based on the systems’ physical
environment and application criticality. The significant objectives of the system
are achieved with the operational requirements of the priority orders. The secu-
rity concept is demonstrated with the system operation in a complex and critical
environment.
7 Challenges in CPS
The development and construction of CPS are generally a complex engineering
process with enormous challenges. Some challenges are not resolved and handled
by the researchers [38]. Those challenges are summarized for analysis, and some
ideas are given for addressing the issues. The models play a substantial role in the
CPS-based process. The heterogeneity is complex with diverse challenges:
• Interoperability of various software systems modeled by multiple people with
diverse technological paradigms, i.e., modeling, language, tools, and theories.
• Spatial-temporal, mobility, and distribution model.
• Synchronization, concurrency, and interaction among the cyber and physical
systems.
236 J. S. Babu et al.
• Validationissues. The CPS applications are mission-critical processes and require
safety measures for real-time, security, concurrent, robustness, and fault tolerance.
The significant factors related to validation issues are:
• Large-scale system with complex nature, i.e., continuous or discrete state.
• Massive security threats over the operational environment and security
vulnerabilities over the cyber systems, network, and communication process.
• The uncertain environment and complex system execution are determined
using the machine learning approaches. Thus, an efficient models or tools
are required to establish effectual verification and validation, fault tolerance,
robustness, security, and so on.
• Evolution issues. The construction of the CPS system cannot be done in a scratch
manner. It is initiated with simple modeling and further evolution. The require-
ments and application domain are dynamically changing. Thus, the CPS is evolved
with a newer and open system model in constant development, and the prevailing
models intend to connect themselves with the CPS. The integration of dynamical
systems over the CPS is a complex task.
• Issues over QoS fulfillment. CPS with various requirements possess various
QoS requirements like maintainability, extensibility, availability, reliability, and
certain other factors. Adaptability, reconfigurability, and evolution deal with the
internal and external uncertainty based on CPS requirements. Additionally, inter-
operability is essential, and some complex applications include a multi-domain
scenario like the smart-health CPS model, pilot system, healthcare system, and
traffic management. The predictability demand is needed to fulfill the CPS
outcomes with essential requirements as all the components are not predictable
like data-driven components. QoS constraints and attributes are modeled for
verification and implementation over diverse CPS applications.
8 Model-Driven Conceptual Approaches
There are enormous techniques, theories, and tools for model-driven (MD)
approaches, including MD engineering, MD- testing, MD architectural model, MD
system engineering, and MD system engineering. These are used for handling various
practical issues and include two concepts such as model transformation and model
modeling [38]. The people construct the modeling phase to examine the subjects and
real-world objects and project the system characteristics. The characteristics of the
modeling phase are given below:
• Abstractive is the concentration toward the specific model with the elimination of
irrelevant models.
• Purposeful construction of the particular set of concerns toward the stakeholders.
• Recognized level of expression to understand the user’s requirements.
• Accurately provisioning the model significance.
• Essential to answer the system model.
A Generic Flow of Cyber-Physical systems—A Comprehensive Survey 237
• It should be faster and cost-efficient for the construction of the existing system.
Based on the available characteristics, the models are partitioned with diverse
categories:
• Construction of the system model satisfying the prescriptive and descriptive levels.
• Including mathematical models like graph theory and so on.
• Categorization of the system model, meta-model, and meta–meta-model.
• Precision is essential with three diverse levels like implementation, conceptual,
and specification models.
The construction of a qualified model is essential, and it needs to model effec-
tual language and offer efficient system design. The modeling specifies the structure
of building models. There are various number of languages like BPMN, AADL,
and UML that have to be proposed. The language modeling is partitioned into
two extensive groups like domain-specific and general-purpose modeling language.
The former model targets supporting certain domains compared to general-purpose
modeling language, which comprises a comprehensive range of modeling abilities
[38]. The domain-specific models are used as the research trends for the construc-
tion of the model. The modeling languages are composed of three diverse compo-
nents: abstract, concrete, and semantics syntax, and it plays a substantial role in
model construction. Various individuals involve themselves in the construction of
large-quality differences and experience the process of guidance (Table 3).
The significant role of model transformation needs to achieve automation with
model-driven approaches [38]. It produces a target model in three diverse forms. (1)
Ta bl e 3 Comparison of various modeling parameters
Categories Remarks
Language-based
properties (nature of
computation)
1.Termination criteria
2.Elimination of execution
semantics
3.Typing
1.Existence of target model
2.Unique target modeling
3.Consistent rule-transformation
4.Concentrating on
transformation language
Transformation-based
properties (nature of
modeling)
1.Model typing and conformance
2.Transformation properties
3.Syntax relationship
4.Semantic relationship
1.Conforms target
meta-modeling
2.Set of the source model
3.Dynamical consistency
4.Mathematical modeling
Testing model 1.Meta-model coverage
2.Direct verification
3.Manual modeling of graph
theories
4.A state-space explosion like
partial order reduction
1.Challenge in defining test
cases
2.Component verification
3.Transformation process
4.Verification transformation
and transformation process
5.Certification by model
verification
238 J. S. Babu et al.
The source and target model consideration is based on the classification of model-to-
model, text-to-model, and model-to-text format. (2) The abstraction level of source
and target models is classified as vertical and horizontal transformation. Most of
the present transformation models are based on meta-model theory only [39]. The
adaptability and scalability of the domain-specific model are provided in a constraint
manner.
9 Conclusion
This work provides an extensive survey on privacy and security measures over CPS
with a specific concentration on various applications like medical devices, smart
grids, ICS, and smart cars. A detailed taxonomy is presented to identify the vulner-
abilities, threats, control methods, and known attacks. Here, a security framework is
integrated with various security and CPS factors. The model projects how the attacks
over the physical domains can influence unexpected consequences over the cyber-
physical fields with the better solutions. Some control mechanisms are modeled to
avoid cyber-physical attacks. For instance, the heterogeneity of the model is identified
with various attacks. Thus, an efficient solution needs to be provided with hetero-
geneous component interaction. The research-based security measures need to be
active with frequent reports over the cyberattacks. However, some defense mecha-
nisms need to be deployed with the system-specific solution and predict vulnerabil-
ities and threats. This survey highlights the challenges over the security mechanism
and hopes that the young researchers intend to handle all these challenges.
References
1. Amin, Schwartz GA, Hussain A (2013) In quest of benchmarking security risks to cyber-
physical systems. IEEE Netw 27(1):19–24
2. Amin, Litrico X, Sastry SS, Bayen AM (2010) Stealthy deception attacks on water SCADA
systems. In: Proceedings of the 13th ACM international conference on Hybrid systems:
computation and control, Stockholm, Sweden, pp 161–170
3. Arias, Wurm J, Hoang K, Jin Y (2015) Privacy and security in internet of things and wearable
devices. IEEE Trans Multi-Scale Comput Syst 1(2):99–109
4. Cao, Zhu P, Lu X, Gurtov A (2013) A layered encryption mechanism for networked critical
infrastructures. IEEE Netw 27(1):12–18
5. Cheminod, Durante L, Valenzano A (2013) Review of security issues in industrial networks.
IEEE Trans Ind Informat 9(1):277–293
6. Cho, Shin KG (2016) Fingerprinting electronic control units for vehicle intrusion detection.
In: Proceedings of 25th USENIX Security Symposium, Austin, TX, USA, 911–927
7. Choi, Kim H, Won D, Kim S (2009) Advanced key-management architecture for secure SCADA
communications. IEEE Trans Power Del 24(3):1154–1163
8. Dardanelli A et al (2013) A security layer for smartphone-to-vehicle communication over
bluetooth. IEEE Embed Syst Lett 5(3):34–37. https://doi.org/10.1109/LES.2013.2264594
A Generic Flow of Cyber-Physical systems—A Comprehensive Survey 239
9. Ericsson (2020) Cyber security and power system communication—essential parts of a smart
grid infrastructure. IEEE Trans Power Del 25(3):1501–1507
10. Fovino, Carcano Am Masera M, Trombetta A (2009) Design and implementation of a secure
modbus protocol. In: Critical infrastructure protection III. Berlin, Germany, Springer, pp 83–96
11. Francia III, Thornton D, Brookshire T (2012) Wireless vulnerability of SCADA systems. In:
Proceedings of the 50th Annual Southeast Regional Conference, Tuscaloosa, AL, USA, pp
331–332
12. Garcia, Oswald D, Kasper T, Pavlidès P (2016) Lock it and still lose it–on the (in) secu-
rity of automotive remote keyless entry systems. In: Proceedings of 25th USENIX Security
Symposium (USENIX Security), Austin, TX, USA, pp 929–944
13. Gollmann (2013) Security for cyber-physical systems. In: Mathematical and engineering
methods in computer science. Berlin, Germany, Springer, pp 12–14
14. Hayajneh, Mohd BJ, Imran M, Almashaqbeh G, Vasilakos AV (2016) Secure authentication
for remote patient monitoring with wireless medical sensor networks. Sensors 16(4):424
15. Jo, Choi W, Na SY, Woo S, Lee DH (2017) Vulnerabilities of android OS-based telematics
system. Wireless Pers Commun 92(4):1511–1530
16. Johnson, “Survey of SCADA security challenges and potential attack vectors,” in Proc. IEEE
Int. Conf. Internet Technol. Secured Trans. (ICITST), London, U.K., 2010, pp. 1–5.
17. Koscher et al (2010) Experimental security analysis of a modern automobile. In: Proceedings
of IEEE symposium on security and privacy (SP), Oakland, CA, USA, pp 447–462
18. Lu, Lu X, Wang W, Wang C (2010) Review and evaluation of security threats on the communi-
cation networks in the smart grid. In: Proceedings of Milcom 2010 Military Communications
Conference, San Jose, CA, USA, pp 1830–1835
19. Mitchell, Chen IR (2013) Behavior-rule based intrusion detection systems for safety critical
smart grid applications. IEEE Trans Smart Grid 4(3):1254–1263
20. Oberoi et al (2016) Wearable security: key derivation for body area sensor networks based on
host movement. In: Proceedings of international industrial electronics (ISIE), pp 1116–1121
21. Alshemali, Kalita J (2019) Toward mitigating adversarial texts. Int J Comput Appl
22. Behjati, S.-M. Moosavi-Dezfooli, M. S. Baghshah, and P. Frossard. Universal adversarial
attacks on text classifiers. In Proc. of IEEE ICASSP, 2019.
23. Oregi, Del Ser J, Perez A, Lozano JA (2018) Adversarial sample crafting for time series
classification with elastic similarity measures. In Proceedings of IDC
24. Ravi, Wong C, Lo B, Yang GZ (2016) A deep learning approach to on-node sensor data analytics
for mobile or wearable devices. IEEE J Biomed Health Inf
25. Ren, Deng Y, He K, Che W (2019) Generating natural language adversarial examples through
probability weighted word saliency. In Proceedings of ACL
26. Zhu, Xiong P, Li G, Zhou W, Philip SY (2018) Differentially private model publishing in cyber
physical systems. Future Gener Comput Syst, in Print
27. Giraldo, Sarkar E, Cardenas AA, Maniatakos M, Kantarcioglu M (2017) Security and privacy
in cyber-physical systems: a survey of surveys. IEEE Design Test 34(4):7–17
28. Gowtham, Ahila SS (2017) Privacy enhanced data communication protocol for wireless
body area network. In: 4th IEEE international conference on advanced computing and
communication systems (ICACCS), 1–5
29. Lu, Zhu H, Liu X, Liu JK, Shao J (2014) Toward efficient and privacy-preserving computing
in big data era. IEEE Netw 28(4):46–50
30. Shim (2016) A survey of public-key cryptographic primitives in wireless sensor networks.
IEEE Commun Surv Tutorials 18(1):577–601
31. Jain, Gyanchandani M, Khare N (2018) Differential privacy: its technological prescriptive
using big data. J Big Data 5(1):15
32. Lv, Zhu S (2019) Achieving correlated differential privacy of big data publication. Comput
Secur 82:184–195
33. Papadimitriou, Li F, Kollios G, Yu PS (2007) Time series compressibility and privacy. In:
Proceedings of the 33rd international conference on Very large data bases. VLDB Endowment,
pp 459–470
240 J. S. Babu et al.
34. Kim, Kumar PR (2012) Cyber–physical systems: a perspective at the centennial. In: Proceed-
ings of the IEEE, 100(Special Centennial Issue):1287–1308
35. Danezis, Fournet C, Kohlweiss M, Zanella-B´eguelin S (2013) Smart meter aggregation via
secret-sharing. In: Proceedings of the first ACM workshop on Smart energy grid security, pp
75–80
36. Gupta (2012) Implantable medical devices-cyber risks and mitigation approaches. In: Proceed-
ings of cybersecurity on cyber-physics and system workshop (NISTIR), pp 15–30
37. Joung (2013) Development of implantable medical devices: From an engineering perspective.
Int Neurourol J 17(3):98–106
38. Rasmussen, Castelluccia C, Heydt-Benjamin TS, Capkun S (2009) Proximity-based access
control for implantable medical devices. In: Proceedings on 16th ACM conference on computer
and communications security (CCS), pp 410–419
39. Rostami, Juels A, Koushanfar F (2013) Heart-to-heart (H2H): authentication for implanted
medical devices. In: Proceedings on ACM SIGSAC conference computer communication
security, pp 1099–1112
Mental Disorder Detection in Social
Networks Using SVM Classification:
An Improvised Approach
B. Dinesh Reddy , Eali Stephen Neal Joshua , N. Thirupathi Rao ,
and Debnath Bhattacharyya
Abstract Online social networking has caused profound changes within the manner
folks communicate and move. These changes might have an effect on sure traditional
aspects of human behavior and cause medical specialty disorders. Mental disease is
quickly turning into one in every of the foremost current public unfitness around
the world. Social media networks, wherever clients will categorical their emotions,
feelings, and thoughts, area unit a worthy supply of knowledge for analyzing mental
state, and procedures supported machine intelligence area unit more and more used
for this purpose. It is difficult to sight social network mental disorder (SNMDs) as
a result of the mental factors thought of in existing diagnostic criteria (question-
naire) cannot be determined from online group action logs. To mechanically sight
SNMDs cases of OSN clients, taking out these constraints to evaluate user’s online
psychological states is extremely difficult. As an instance, the range of isolation with
the impact of less inhibition of OSN clients do not seem to be simply discernible.
Abnormal activity connected keywords area unit generated and hold on server. Each
user activity (tweets, post, comments, etc.) information area unit hold on in infor-
mation that may be accustomed analyze folie. This can facilitate to observe user
activities in social network. Projected work detects sever style of SNMDs with a
binary SVM classification approach.
Keywords Mental disorder ·Data mining ·Online social network ·SVM ·SNMD
B. D. Reddy (B) · E. S. N. Joshua · N. T. Rao
Department of Computer Science and Engineering, Vignan’s Institute of Information Technology,
Visakhapatnam, AP 530016, India
e-mail: dinesh4net@gmail.com
D. Bhattacharyya
Department of Computer Science and Engineering, Koneru Laksmaiah Education Foundation,
Vaddeswaram, Guntur, AP 522302, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_25
241
242 B. D. Reddy et al.
Fig. 1 Showing the proposed data flow design for decision-making
1 Introduction
The important objective of the information mining technique [1] is to take out data
from a data set, rework it to an evident structure for huge usage apart from the
raw analysis method, it involves information data, knowledge management aspects,
knowledge pre-processing, design and logical thinking issues, powerfulness metrics,
quality issues, post-processing of discovered structures, database, knowledge repos-
itory, Web, data files with different papers are the particular origin of knowledge. We
would like giant huge amount of existing knowledge for facts processing in order
to achieve success. Institutions typically stock knowledge in the knowledge bases
or data repositories. Knowledge repositories might have one or a lot of data files,
worksheets, or another forms of data repositories (Fig. 1).
In some periods, knowledge might be residing on even in plain data files or
worksheets. Internet or World Wide Web is the massive supply of knowledge.
2 Literature Survey
We reveal that these representations could also be simply value-added to the previous
models and considerably upgrade the state of the art across six difficult information
science issues, alongside respondent [2–4], matter deduction, and sentiment analysis.
Mental Disorder Detection in Social Networks … 243
We have a bent to additionally gift associate in Nursing examining revealing that
uncovering the deep down of the previously-trained network is critical, permitting
down river design to mingle differing kinds of partial-supervision signals. Combining
inner states during this way permits for very made word representations. Exploitation
intrinsic evaluations we have a bent to point out the greater-level LSTM [5] states
catch context-dependent aspects of signified (example, they are going to be utilized
while not alteration to function well on supervised knowledge illumination tasks),
whereas smaller level states design features of structure (example, they are going to
be accustomed do POS-tagging).
At an equivalent time, exposing all of these signals extremely useful, permitting
the experienced design choose the categories of partial-supervision that the area unit
most helpful for every finish task. Since addition of ELMo upgrades task functional-
ities over word vectors alone, the biLM’s discourse representations should inscribe
info usually helpful for information processing tasks that is not captured in word
vectors.
Suggest universal language model fine-tuning (ULMFiT) [6], an honest shift
learning methodology will be applicable to any function in NLP and introduce
methods that are used for fine-tuning language design. Notwithstanding however,
numerous the general-domain information used for pre-training is, the info of the
target task can doubtless return from a unique distribution. We have a tendency
[7] to s o fine-tune the lumen on information of the target task. Given a pre-trained
general-domain lumen, this stage converges quicker because it solely has to adapt to
the idiosyncrasies of the target information, and it permits U.S.A. to coach a sturdy
lumen even for little data sets. For adapting its parameters to task-specific options,
we had just like the model to quickly converge to an appropriate region of the param-
eter house within the starting of coaching so refine its parameters. Victimization
[8] constant learning rate (LR) associates degree treated learning rate throughout
coaching is not the simplest thanks to attain this behavior. Instead, we have a tendency
to propose slanted triangular learning rates (STLR) that 1st linearly will increase the
training rate so linearly decays it consistent with the subsequent update schedule.
3 Existing System
A social media service (also social networking Internet location) may be a net plat-
form, employed by individuals to make social networks or social relations with
people. The variety of full and essential social media services presently offered within
the online shows consequences of definition; but, there are some general features:
(1) social media services are Internet-based applications, (2) user-generated content
(UGC) is that the life of SNS creatures, (3) users produce service-oriented profiles
for the location or applications that are modeled and maintained by the SNS institu-
tion, and (4) social media services make easier the event of online social networks by
connecting a client’s profile with those of other people and teams. Most social media
services are Internet-based and provide facility for clients to act in the online, like by
244 B. D. Reddy et al.
e-mail (g-mail) and instant messaging apps and online sites. Social media applica-
tions are differed which they include/absorb a spread of recent data and communica-
tion tools like handiness on personal computers/desktop and laptops, mobile phones
like tablet. Online community services are generally thought-about a social media
service, tho’ during a bigger sense, social media service typically signifies associate
individual-oriented service whereas online community services are group-oriented.
Social media sites enable clients to disclose concepts, digital photos and videos,
posts, and intimate others regarding online or other planet functions and events with
individuals in their network. Whereas individual social Internet like collecting during
a goods market in village to speak regarding functions has existed during the earliest
modeling of cities, Internet allows individuals to connect with others World Health
Organization board totally various locations, scaling from in between a town to in
between the earth. Depending on the social media Web pages, associates could even
be ready to contact the opposite member. On alternative situations, associates will
connect to any member they have an association with, and afterward any member that
connect incorporates an association with, and so on. LinkedIn, a career social media
service usually needs that an associate face to face understand another associate in
the world before they connect to them online. Certain services need associates to be
having an antecedent association for connecting alternative associates.
4 Proposed System
The important sorts of social media services area unit that those hold class areas
(such as previous school year or classmates) signify pointer-connect with compan-
ions (normally with own- described pages), and a proposed system joined to believe.
Social media services are going to be divided into 3 kinds: socialization social media
services area unit initially for socialization with the previous friends (example);
online social media services area unit initially for self- contained social interaction
(example, LinkedIn, a future career and employment-oriented page); and social navi-
gable social media services area unit initially for serving to clients to hunt out certain
data or assets. They are making an effort to qualifying these services to exclude the
need to corresponding entries of companions and interests. People area unit progres-
sively exploitation social media Web pages, like Twitter and Facebook, to disclose
their comments and views with their connections. Sharing posts on these pages area
unit related during a very representational connections, within the period of daily
actions and acts that happens. Like that, social media gives a way for catching activity
constraints which area unit similar to Associate in Nursing individual person’s
thinking, mindset, interaction, actions, and socialism. The sensation and vocabulary
utilized in social network sharing’s might represent the depressed feel of uselessness,
self-guilt, vulnerableness, and shame that differentiates greater depression. To boot,
despondency sufferers usually pull out from communal things and activities. Such
differences in activity could be notable with differences in activity on social network.
And also, Internet pages might mirror ever-differing social interactions. We tend to
Mental Disorder Detection in Social Networks … 245
Fig. 2 Showing the system architecture of the proposed model
follow the supposition that differ in vocabulary, activity, and social ties could even be
utilized collectively to develop applied mathematics designs to sight and even find
major depressive disorder during a very close-grained method, alongside approaches
during which will complete and expand ancient ways to identification (Fig. 2).
SNSs permit people to design a public profile, tons of or slightly visible in
step with inaction of Website and user wisdom, produce an inventory of alternate
clients with United Nations agency which move then consider record of contacts
created by alternative clients at intervals of situation. On social network services,
every client will relate themselves, stepping into data concerning of their back-
ground details (example, high school), demographics (example, sex, age group),
and cultural behavior (example, books of favorite, films, TV programs); clients
will decide pictures and conjointly add on their selves on their profiles, creating
self-description. Planned work uses data processing techniques to observe SNMDs
[9, 10].
Text mining is employed to investigate the text and classify supported abnormal
keyword that keep on info. Propose associate degree approach, new this apply of
SNMDs detection, by analyzing information logs of online social network clients to
246 B. D. Reddy et al.
earnestly determine powerful social network mental disorder cases early. Use ML
structure for detection of SNMDs. Planned pattern is formed with (SVM) that has
been wide accustomed analyzes online social networks in several areas.
5 Methodology
5.1 OSN Framework Construction
Social network indicates the communication among people during when they make,
transfer, or share/disclose data and theories in virtual groups and networks. Product,
individual, or corporation during this module, we are able to have 3 processes like
OSN users, knowledge analysis, disorder prediction.
5.2 Data Collection
Data sets square measure collected supported obtained from social networks. Here
known two broad approaches to knowledge assortment: (1) grouping knowledge
directly from the participants with their consent mistreatment surveys and electronic
knowledge collection instruments and (2) aggregating knowledge extracted from
public posts (Fig. 3).
5.3 Pre-Processing
The collected information is typically pre-processed by (1) removing irrelevant
samples and (2) cleanup and preparing the data for analysis. Pre-processing is
employed to get rid of incomplete details, that area unit typically removed so as
to boost the accuracy of prediction and classification results. Every post was pre-
processed by eliminating the stop words and suitable information (e.g., retweets,
hashtags, uniform resource locators), lowercasing characters, and segmenting
sentences (Fig. 4).
5.4 Feature Extraction
In text data mining, mentality analysis may be a well-liked tool for comprehension
of feeling expression. It is used to differentiate the contrariness of a given text into
classes like positive, non- positive, and impartial. Feature choice isolates a relevant
Mental Disorder Detection in Social Networks … 247
Fig. 3 Showing the OSN framework constructions
Fig. 4 Showing the framework of the model
248 B. D. Reddy et al.
Fig. 5 Showing the pre-processing of the model
set of options that area unit ready to know symptoms of mental instability or properly
label the participants, whereas exclude overfitting. Applied mathematics analyzing
is usually done to find the group of attributes which may differ between users with
mental instability and users while not mental instability (Fig. 5).
Prediction Model
Prediction models were accustomed sight and classify users in line with mental
instability and gratification with life. To make a prophetic design, specific list of
options will be employed as a coaching knowledge for ML algorithms to find out
sequence from that knowledge [11]. Planned structure was made by support vector
machine (SVM) that has been wide accustomed analyzes OSNs in several areas.
Result
See Fig. 6.
6 Conclusion
The input style is to gather social network information, from user feed information on
OSN framework. Information was cleansed and pre-processed to confirm that they
Mental Disorder Detection in Social Networks … 249
Fig. 6 Showing the performance of the proposed model
are within a kind needed by the analyzing algorithms. Then, main options associated
with the analysis field are ready for model building. Overall, this includes feature
extraction and has choice, manufacturing list of options to utilize in understanding
and supportive prophetic models. The projected style is for prediction of the mental
stress gift or not. Feature choice separates a similar set of options from which area
unit able to predict causes of moving stress which will differ from clients with mental
disabilities and clients while not mental disabilities.
7 Future Work
In future gift, a structure for sleuthing client’s mental stress states from client’s weekly
social Website knowledge, investment tweet’s contents in addition as client’s social
communications using real-world social media knowledge because t he basis, studied
the correlation between client’s mental stress states and their social communication
activities.
250 B. D. Reddy et al.
References
1. Bhattacharyya D, Kumari NMJ, Joshua ESN, Rao NT (2020) Advanced empirical studies on
group governance of the novel corona virus, mers, sars and ebola: a systematic study. Int J
Current Res Rev 12(18):35–41. https://doi.org/10.31782/IJCRR.2020.121828
2. Doppala BP, NagaMallik Raj S, Stephen Neal Joshua E, Thirupathi Rao N (2021) Automatic
determination of harassment in social network using machine learning. https://doi.org/10.1007/
978-981-16-1773-7_20, Retrieved from www.scopus.com
3. Eali SNJ, Bhattacharyya D, Nakka TR, Hong S (2022) A novel approach in bio-medical image
segmentation for analyzing brain cancer images with U-NET semantic segmentation and TPLD
models using SVM. Traitement Du Signal 39(2):419–430. https://doi.org/10.18280/ts.390203
4. Eali SNJ, Rao NT, Swathi K, Satyanarayana KV, Bhattacharyya D, Kim T (2018) Simulated
studies on the performance of intelligent transportation system using vehicular networks. Int J
Grid Distribut Comput 11(4):27–36. https://doi.org/10.14257/ijgdc.2018.11.4.03
5. Joshua ESN, Battacharyya D, Doppala BP, Chakkravarthy M (2022) Extensive statistical anal-
ysis on novel coronavirus: towards worldwide health using apache spark. https://doi.org/10.
1007/978-3-030-72752-9_8. Retrieved from www.scopus.com
6. Joshua ESN, Bhattacharyya D, Chakkravarthy M (2021) Lung nodule semantic segmentation
with bi-direction features using U-INET. J Med Pharm Allied Sci 10(5):3494–3499. https://
doi.org/10.22270/jmpas.V10I5.1454
7. Joshua ESN, Bhattacharyya D, Chakkravarthy M, Kim H (2021) Lung cancer classification
using squeeze and excitation convolutional neural networks with grad cam++ class activation
function. Traitement Du Signal 38(4):1103–1112. https://doi.org/10.18280/ts.380421
8. Joshua ESN, Chakkravarthy M, Bhattacharyya D (2021_ Lung cancer detection using impro-
vised grad-cam++ with 3D CNN class activation. https://doi.org/10.1007/978-981-16-177
3-7_5, Retrieved from www.scopus.com
9. Bhattacharyya D, Doppala BP, Thirupathi Rao N (2020) Prediction and forecasting of persistent
kidney problems using machine learning algorithms. Int J Current Res Rev 12(20):134–139.
https://doi.org/10.31782/IJCRR.2020.122031
10. Neal Joshua ES, Bhattacharyya D, Chakkravarthy M, Byun Y (2021) 3D CNN with visual
insights for early detection of lung cancer using gradient-weighted class activation. J Healthcare
Eng. https://doi.org/10.1155/2021/6695518
11. Neal Joshua ES, Chakkravarthy M, Bhattacharyya D (2020) An extensive review on lung
cancer detection using machine learning techniques: a systematic study. Revue d’Intelligence
Artificielle 34(3):351–359. https://doi.org/10.18280/ria.340314
An Enhanced K-Means Clustering
Algorithm to Improve the Accuracy
of Clustering Using Centroid
Identification Based on Compactness
Factor
Eali Stephen Neal Joshua , K. Asish Vardhan, N. Thirupathi Rao ,
and Debnath Bhattacharyya
Abstract The researchers find it difficult to extract information from a large data
set through a standard function. It is found insufficient of standard functions to
extract needed information. It has been considered that the k-means algorithm in
the situation where the data is too enormous to be stored in main memory and must
be retrieved sequentially, such as from a disk, and where it must be used as slight
memory as possible. The k-means clustering also converges very quickly when it is
employed to obtain data from huge data collections. It is also on other hand, k-means
has some disadvantages too, and it includes affluent computation by getting cluster
centers which are randomly selected at initial. It influences the two factors, such as
performance of the algorithm and number of clusters initialization. In this paper, an
improved k-means algorithm in terms of data clash strainer mechanism is given. The
data clash strainer mechanism is implemented through a function regional centroid
component (RCC) mechanism which is added to the standard k-means algorithm.
This density-based recognition mechanism is built on the properties of clash data.
The clustering result is effectively enhanced by ignoring the clash data prior to the
process of data clustering. Hence, the improved algorithm offers a great accuracy
when compared to other existing cluster algorithms.
Keywords Cluster ·Data clash strainer ·K-means ·RCC
E. S. N. Joshua (B) · N. T. Rao
Department of Computer Science and Engineering, Vignan’s Institute of Information Technology,
Visakhapatnam, AP 530016, India
e-mail: stephen.eali@gmail.com
K. A. Vardhan
Department of Computer Science and Engineering, Bullayya College of Engineering for Women,
Visakhapatnam, AP, India
D. Bhattacharyya
Department of Computer Science and Engineering, Koneru Laksmaiah Education Foundation,
Vaddeswaram, Guntur, AP 522302, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_26
251
252 E. S. N. Joshua et al.
1 Introduction
A vast amount of data is dealt in various fields, and those big data is handled using
data mining techniques to retrieve information. “We are living in the information age”
is a popular saying; however, it is like actually living in the data age. Terabytes or
petabytes of data pour into our computer networks, the World Wide Web, and various
data storage devices every day from business and which is needed to be extracted in a
useful manner to infer knowledge from it [1]. The technique of data mining involves
the cluster analysis which is one of the main focuses of the present-day researchers.
Clustering is a fundamental method for appreciative and interpreting data that
seeks to partition input objects into groups, known as clusters, such that objects
within a cluster are similar to each other, and objects in different clusters [2]are
not. A clustering invention called k-means is simple, intuitive, and widely used in
practice. Given a set of points S in a Euclidean space and a parameter k, the objective
of k-means is to partition S into k clusters in a way that minimizes the sum of the
squared distance from each point to its cluster center [3]. This circumstance causes
the formation of wide range of clustering algorithms such as COBWEB, DBSCAN,
CURE, and MEANS [4].
This work introduces the method that avoids the arbitrary selection of options
at initial and involves detection and eliminations of the identified far-apart data
collection from the clusters. It works initially to improve the performance of classic
k-means clustering mechanism in terms of its accuracy and reduced complexity.
The rest of the paper is organized as follows; the Section “Related Works”gives
the brief discussion on related works. In Section “Centroid Recognition Based on
Compactness”, the basic nature of k-means clustering procedure is studied, and
proposed methodology based on the compactness-based centroid detection tech-
nique is presented. Section “Results and Discussion” focuses on the comparison of
proposed methodology with other existing clustering algorithms and gives the results
of experiments. Finally, Section “Conclusion” concludes the current work.
2 Related Works
Shorab et al. [5] present an empirical method to select the appropriate centroids at
initial level in k-means clustering strategy, and hence, it tries to improve the algorithm
in terms of its clustering accuracy as well as focused on the time of execution.
Experimental results showed the better adeptness of the improved k-means clustering
algorithm over the traditional k-means algorithm but it increases complexity of the
clustering algorithm as the size of data set increases.
Cosmin et al. [6] reveal customer segmentation which is done with data mining
to know the customer characteristics information hidden inside. The way to find out
the customer segments of a company is clustering analysis. Clustering is the process
An Enhanced K-Means Clustering Algorithm to Improve the Accuracy … 253
of forming segments of a set of data by measuring similarities between data with
other data.
Patel and Prateek [7] explore different kind of various problems using data mining
clustering mechanism and the relationships between them. K-means clustering algo-
rithms, hierarchical algorithm is discussed in this paper. The performance of this
algorithm is compared in clustering process and gives the proposition about the
suitability of such algorithms in different kind of states for the different data sets.
Syakur et al. [8] say that the segmentation process puts customers in line with the
characteristics of similar customer groups. Customer segmentation is a preparatory
step to classify each customer according to a defined customer group. Customer
segmentation based on market research and demography requires understanding the
characteristics of all customers to be more effective.
Hong et al. [4] proposed an improved k-means algorithm as the result of clustering
reliability analysis, and the proposed algorithm shows the stability and achieves better
result when the solidity is uneven, and there exist large difference in data clustering.
Experimental results showed the ability of improved k-means algorithm in handling
non-uniformed data set.
Proposed Methodology
The result of cluster analysis based on partition strategy the k-means methodology
was derived [9]. This methodology requires arbitrary selection of “k” number of
cluster centroids at initial. It also involves computation of distance between each
selected centroid and each instance of organized data collection to find the nearest
centroid and also amend average distance of centroids. This process is repeated until
standards or norms of the function met.
The mean squared deviation standard for clustering is calculated as follows,
Where lij is instance of class I and si is centroid of class i. This methodology
is illustrated in the Fig. 1. This k-means clustering algorithm steps were given as
follows. It involves the arbitrary selection of centroids, detection of data center point,
calculating distance, forming clusters.
Input: P instances need to be cluster {a1, a2, ……, an} and the k (no. of initial
centroids).
Output: k centroids and the disagreement volume between each instance and its
short-distant centroid neighbor.
The intricacy of this k-means mechanism is expressed by the factors—arbitrary
selected k number of clusters, number of repetitions of the procedure, and number
of organized data instances [10].
3 Centroid Recognition Based on Compactness
The performance of the k-mean technique depends on the centroid selected at initial
which greatly affects the result of the technique used. The outliers in the clusters
away from the data-compact region cause the newly founded centroid more deviated
254 E. S. N. Joshua et al.
Fig. 1 Showing the performance of the proposed classifiers
from data-compact region and so, it directly influences the final clustering result,
and the final result encounters the huge deviation from the actual. To avoid such
an outlier and to enhance the result, it is better to discard the isolated data from
our collection of data prior to the process of data clustering. The deviating level of
each instance in organized data is determined using regional centroid component
(RCC) which involves the computation of distance of each instance from its short-
distant centroid neighbor only after the completion of the process of producing k
number of centroids and k number of shortest distance of each instance from its
short-distant centroid neighbor. Finally, RCC detects the regional centroid as per the
regional centroid component of each instance [13]. This regional centroid component
(RCC) detection is illustrated in the following steps: Where SDA (i) is the regional
compactness of k-short-distant centroid of d, and SDA (d) is regional compactness
of d. RCC (d) expresses the scope of d as centroid. The RCC has the value of about
one in compactness dispensation data collection. The centroid component by which
the centroid is differentiated is greater than others because the regional compactness
of the centroids in the collection of data is much less than the regional compactness
of its short-distant instances.
3.1 The Improved k-means Clustering Algorithm Using
Regional Centroid Component (RCC)
The mechanism is triggered by the process of elimination of far-apart data collection
by employing the above said RCC-based recognition strategy. It ensures that the
An Enhanced K-Means Clustering Algorithm to Improve the Accuracy … 255
computation of initial centroid is free from the far-apart data collection instances and
removes them in the determination of centroid. The improved k-means algorithm is
executed on the newly selected data organization employing the RCC and illustrated
in the following steps.
Input: P instances need to be cluster {a1, a2, ……, an} and the k (no. of initial
centroids).
Output: k centroids and the disagreement volume between each instance and its
short-distant.
4 Results and Discussion
The performance of the proposed methodology in terms of its suitability and accuracy
by comparing it with other existing clustering algorithms like mean shift clustering,
density-based spatial clustering application noise (DBSCAN), expectation–maxi-
mization (EM) clustering using Gaussian mixture models (GMMs), and agglomer-
ative hierarchical clustering. The data sets from the UCI—one of the most popular
neural network database—Abalone, Wine, and Iris have been taken for our exper-
iment. Table 1 gives these details in brief. The proposed work produces the better
outcomes and offers optimal solution without avoiding the caliber of clustering. The
experimental results prove the effectiveness of this improved version of k-means
algorithm over all of other clustering strategies.
The accuracy level of the proposed work along with mechanism, and it had
achieved greater accuracy due to the introduction of component strategy in traditional
k-means algorithm.
From the Fig. 5, it can be clearly noted that the accuracy level of the proposed
RCC-based improved k-means provides approximately10% higher level of accuracy
when compared to other existing clustering algorithms. Furthermore, our proposed
algorithm exhibits stability and parallelization efficiency and proves its greater
usability.
Clustering time comparison is also made with all other clustering strategies.
Although the proposed algorithm’s clustering time is greater than others, the differ-
ence in the time taken of other algorithms is not a considerable amount, and proposed
method’s consumption closely resembles the others.
Table 1 Selected data sets Data set No. of objects No. of properties No. of groups
Abalone 15,100 11 36
Iris 1400 24 6
Wine 3462 42 7
256 E. S. N. Joshua et al.
5 Conclusion
The arbitrary selection of “k” number of initial options in classical k-means algorithm
makes it instable and increases complexity of the algorithm, and hence, its overall
performance gets reduced in terms of accuracy. The problem is been overcome by
introducing a novel compactness-based regional centroid component recognition
(RCC) function into the k-means clustering technique. As illustrated in experimental
results, it has been achieved the better accuracy than all of the other existing clustering
algorithms in almost same amount of time taken by other algorithms.
References
1. Satyanarayana KV, Rao NT, Bhattacharyya D, Hu Y (2022) Identifying the presence of
bacteria on digital images by using asymmetric distribution with k-means clustering algorithm.
Multidimension Syst Signal Process 33(2):301–326. https://doi.org/10.1007/s11045-021-008
00-0
2. Birant D, Kut A (2007) ST-DBSCAN: an algorithm for clustering spatial-temporal data. Data
Knowl Eng 60(1):208–221
3. Miller KG, Lee RP, Tableman A et al (2021) Dynamic load balancing with enhanced shared-
memory parallelism for particle-in-cell codes. Comput Phys Commun 259, Article ID 107633
4. Bhattacharyya D, Doppala BP, Thirupathi Rao N (2020) Prediction and forecasting of persistent
kidney problems using machine learning algorithms. Int J Current Res Rev 12(20):134–139.
https://doi.org/10.31782/IJCRR.2020.122031
5. Zhou Q, Guo S, Lu H (2020) Falcon: addressing stragglers in heterogeneous parameter server
via multiple parallelism. IEEE Trans Comput 70(1):139–155
6. Shibao S, Keyun Q (2007) Research on modified k-means data cluster algorithm. IS Jacobs,
Bean CP. Fine particles thin films and exchange anisotropy. Comput Eng 33(13):200–201
7. Vardhan KA, Rao NT, Raj SNM, Sudeepthi G, Divya, Bhattacharyya D, Kim T (2019) Health
advisory system using IoT technology. Int J Recent Technol Eng 7(6):183–187. Retrieved from
www.scopus.com
8. ALN Fred, JMN Leitão (2000) Partitional vs hierarchical clustering using a minimum grammar
complexity approach. Proceedings of the SSPR and SPR 2000. LNCS 1876, pp 193–202.
[online] Available: http://www.sigmod.org/dblp/db/conf/sspr/sspr2000.htm
9. Fahim AM, Salem AM, Torkey FA (2006) An efficient enhanced k-means clustering algorithm.
J Zhejiang Univ Sci A 10:1626–1633
10. Yuan F, Meng ZH, Zhang HX, Dong CR (2004) A new algorithm to get the initial centroids.
Proceedings of the 3rd international conference on machine learning and cybernetics, pp 26–29
Prediction of Chronic Kidney Disease
with Various Machine Learning
Techniques: A Comparative Study
K. Swathi and G. Vamsi Krishna
Abstract Chronic kidney disease is one of the serious health care issues faced by
people across the globe. It is majorly resulting in kidney failure or sometimes leads
to cardiovascular disease, or sometimes leads to the death of a person. So, the detec-
tion of this disease in the early stages plays a significant role which helps in treating
and controlling the disease. In this paper, various machine learning algorithms are
demonstrated that disclose and extract hidden information from clinical and labora-
tory patient data, which can aid clinicians in maximizing accuracy for illness severity
stage assessment. Several machine learning algorithms like KNN, RF, AdaBoost,
gradient boost, and a voting classifier were considered, and a comparative study was
done. These comparisons were made by taking the CKD dataset available in the UCI
repository. The models employed for the study provide much accuracy, greater than
prior research, suggesting that they are more trustworthy than the previous models.
Keywords Classification ·Machine learning ·Chronic kidney disease
1 Introduction
Chronic kidney disease (CKD) is a long-lasting disease that affects the kidney that
may further lead to end-stage renal failure, which will stop the entire kidney from
functioning and not be able to perform the waste removal or excess water or any
chemicals from your body may cause disparity [1]. The renal failure might expose
cardiac arrest and various artery failures and lead to death. CKD affects various
people worldwide and ranges between 7 and 15%. Globally, in 2007, around 1.21
K. Swathi (B)
Department of Computer Science and Engineering, Vignan’s Institute of Information Technology
(A), Visakhapatnam, AP, India
e-mail: swathi.kalam@gmail.com
G. Vamsi Krishna
Department of Computer Science and Engineering, Dr. Lankapalli Bullayya College of
Engineering, Visakhapatnam, AP, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_27
257
258 K. Swathi and G. Vamsi Krishna
million people were found dead because of this CKD, and the mortality rate increased
in the next years [2].
Early detection of kidney damage can help with treatment, which is not always
possible. To avoid serious injury, we need to comprehend a few renal illness indica-
tions better. The major motive of this examination is to anticipate kidney sickness by
undergoing a deep analysis of data from the observed indices, using various machine
learning classification algorithms to prognosticate the ill health, and then predicting
it by selecting the best technique that gives a better accuracy estimate [3].
The main objective of this work is to undertake a comparative examination
of multiple machine learning algorithms to predict renal illness. The accuracy
percentage for most of the studies was above 90%, which was regarded as excel-
lent. This paper is unique because it employs a variety of algorithms and achieves
above 97% accuracy rate, which is greater than in the previous studies.
2 Literature Review
Many researchers are working on the prediction of CKD using a variety of clas-
sification algorithms. They evaluate the algorithms, namely random forest and the
back propagation approach, and found that back propagation, a supervised learning
model called a feedforward neural network, produces the best results [4]. Finally,
for the system, the random forest implementation method is selected [5]. W.H.S. D
Gunarathne et al. [6] made the comparison to find out the solution from different
machine learning models. Out of all algorithms, they finally found that the multi-
class decision forest technique performs best compared with earlier techniques, with
a higher accuracy rate of 99%. However, this algorithm works well when the smaller
dataset is taken, and they take only 14 attributes. Ramya et al. [4] employed multiple
machine learning classification algorithms to minimize the diagnostic meter and to
upgrade diagnostic accuracy rate. Reddy et al. [7] tested 12 different classification
algorithms on a CKD dataset of 400 records and 24 attributes. They assessed the
accuracy of prediction findings by comparing computed to actual results. As evalu-
ation criteria, accuracy which will tell how well a model works, sensitivity tells how
good a model can identify true positive instances, precision, and specificity were
used. The decision tree approach has an accuracy of up to 98.6%, a sensitivity of
0.9720, a precision of one, and a specificity of one. Arif-Ul-Islam et al. suggested a
method that uses boosting classifiers and J48 decision tree to forecast sickness. This
present work aims to identify the CKD by examining the boosting algorithm perfor-
mance. They also derive the regulations that illustrate correlations between CKD
features. The model’s accuracy will be based on the prediction of outputs and will
be affected by missing values in the dataset. They found a solution to this problem
by recalculating CKD stages, which resulted in uncertain results. To fill up the gaps,
they computed missing data and use a machine learning paperback method to detect
CKD. They get their data from a 400-record dataset with 25 factors that indicate
Prediction of Chronic Kidney Disease with Various Machine Learning … 259
Figure 1
Gradient
Boosting
ADA Boost
Algorithm
Data set
Data pre-processing
Feature selection
Model Train
Model Test
Fig. 1 Block diagram
whether or not a patient has CKD. They use K-nearest neighbors, neural networks,
and random forest to arrive at their conclusions.
3 Materials and Method
This section provides block diagrams, flow diagrams, evaluation matrices, and the
study’s approach and methodology, as well as a description of the dataset.
The suggested system is represented in Fig. 1 by a block diagram. The CKD
prediction dataset is used by the framework. Gradient boosting, KNN, AdaBoost
algorithm, and random forest algorithms have all been employed after pre-processing
and feature selection. In the next subsections, we will go over each of the diagram’s
components in detail.
Dataset: The CKD dataset was used for this study. This data collection consists
of 400 rows and 20 columns. A value of “1” or “0” appears in the output column
“class.”
Feature Selection: It is a method of selection of only the necessary features that
are needed for our model training.
4 Results and Discussion
1. Gradient Boosting: It is one of a kind of ensemble technique and the most
powerful algorithm to deal with tabular data. We can even find from the complex
260 K. Swathi and G. Vamsi Krishna
problems. Usually, the complex problems do have non-linearity and that can be
used predicted using the non-linear activation function like ReLU, Sigmoid, etc.
This algorithm will also help you to deal with missing values. We can achieve
better performance by combining together multiple models which are weak. It
can have one or more functions as illustrated by the gradient function. And this
also cut down the loss functions by continuously repeating the function over a
data point again and again. This works on improvement of loss function, and
it is determined as a weak learner. It performs randomized sampling of data.
It can reduce overfitting of data so the model performance can be increased. It
uses sequential classifier because it is a boosting technique. When applied on the
dataset using gradient boost, the accuracy achieved is 97.8% and is shown with
the help of ROC curve.
2. AdaBoost Algorithm: Versatile helping likewise know as AdaBoost takes addi-
tional duplicates of a base classifier continuously on the equivalent dataset.
Choice stumps are utilized as feeble students. Choice stumps are only trees
which have just a single split. More weight is given to hard to characterize
occasions though lesser weight is given simple to arrange perceptions. A normal
of the weighted yield from every one of the singular students gives the eventual
outcome. Using this AdaBoost algorithm, the accuracy obtained on the dataset
is 95.6% and the ROC curve as follows (Figs. 2 and 3):
3. K-nearest Neighbor
K-nearest neighbor is the semi-parametric artificial intelligence estimations
considering the controlled learning strategy’s NN computation, and it uses the
immediacy to exhibit the classifications determining how the grouping of similar
objects. The KNN can be used to work with classification as well as regression
problems. In view of classification of different classes is based on majority of
Fig. 2 ROC curve for test data using gradient boosting algorithm
Prediction of Chronic Kidney Disease with Various Machine Learning … 261
Fig. 3 ROC curve for test data using AdaBoost algorithm
votes. And the regression is based on calculating Euclidean distance, which is
used to find out the distance between two nearest data points, and the objects are
clustered which are of nearest distance. The KNN algorithm is regarded as a lazy
because it can only have the capability to store the training data, and whenever
any classification is made it creates over head on the memory pool where the
training data is residing. While storing the data in memory, it will not perform
any sort of calculations. It always tries to find out the points to determine to which
cluster that particular data point refers. And it is a simple way to classify the data.
If any newly discovered data comes into the account, the classifier will classify
that data into a cluster to where it should belong to the kidney dataset uses KNN
algorithm which results in 91.3% accuracy, and ROC is shown as Fig. 4:
5 Conclusion
This research aims to observe and examine the outcomes obtained by employing
various AI computations to predict chronic kidney failure in the clinical area. This
paper presented an expectation computation to predict CKD in its early stages. The
dataset includes input limitations gathered from CKD patients, and the models are
ready and authorized for the specified data limitations. Gradient boosting, AdaBoost,
KNN, and random forest learning models are used to complete CKD. The models’
performance is evaluated in terms of assumption accuracy. The assessment findings
revealed that the gradient supporting model better predicts CKD in conjunction with
AdaBoost, random forests, and KNN. The comparison should also be conceivable
based on the execution time, including setting the decision as to the act of spontaneity
of this investigation. A combination of classifiers is also implemented as part of voting
262 K. Swathi and G. Vamsi Krishna
Fig. 4 ROC curve for test data using K-nearest neighbor algorithm
classifiers, where voting classifier 1 uses RF, KNN, and gradient boost, resulting in
the highest accuracy equal to the gradient boost algorithm, whereas voting classifier
2 uses RF and KNN, resulting in lower accuracy of 95.6% when compared to voting
classifier 1 (GB versus KNN versus RF).
References
1. Hooi LS, Ong LM, Ahmad G, Bavanandan S, Ahmad NA, Naidu BM et al (2013) A population-
based study measuring the prevalence of chronic kidney disease among adults in West Malaysia.
Kidney Int 84(5):1034–1040. pmid:23760287
2. Chandra Sekhar, P., Thirupathi Rao, N., Bhattacharyya, D., & Kim, T. -. (2021). Segmentation
of natural images with k-means and hierarchical algorithm based on mixture of pearson distri-
butions. Journal of Scientific and Industrial Research, 80(8), 707–715. Retrieved from www.sco
pus.com
3. Gao S, Manns BJ, Culleton BF, Tonelli M, Quan H, Crowshoe L et al (2007) Prevalence of
chronic kidney disease and survival among aboriginal people. J Am Soc Nephrol 18(11):2953–9.
pmid:17942955
4. Eali SNJ, Bhattacharyya D, Nakka TR, Hong S (2022) A novel approach in bio-medical image
segmentation for analyzing brain cancer images with U-NET semantic segmentation and TPLD
models using SVM. Traitement Du Signal 39(2):419–430. https://doi.org/10.18280/ts.390203
5. Bhattacharyya D, Doppala BP, Thirupathi Rao N (2020) Prediction and forecasting of persistent
kidney problems using machine learning algorithms. Int J Curr Res Rev 12(20):134–139. https://
doi.org/10.31782/IJCRR.2020.122031
6. Doppala BP, Raj SNM, Joshua ESN, Rao NT (2021) Automatic determination of harass-
ment in social network using machine learning. https://doi.org/10.1007/978-981-16-1773-7_20
Retrieved from www.scopus.com
7. Reddy DKK, Behera HS, Nayak J, Routray AR, Kumar PS, Ghosh U (2022) A fog-based
intelligent secured IoMT framework for early diabetes prediction, 199–218
Blockchain and Its Idiosyncratic Effects
on Energy Consumption
and Conservation
K. Mrudula Devi, D. Surya Sai, N. Thirupathi Rao ,K.Swathi ,
and Swathi Voddi
Abstract Blockchain technology, which has affected almost every industry and the
business model, takes no exemption of the vast energy sector. Energy businesses in the
world have already started reconnoitering the use of blockchain technology in various
applications. The applications range from P2P energy trading, asset management,
and demand and supply chain tracking. Moreover, with precise reliability alongside
security requirements, the application range of the energy sector is relatively narrow.
The article aims to define and show the environment and methodology for applying
blockchain principles for solving operational technology challenges at energy utili-
ties and for a much closer analysis of all the possible integrations between the energy
sector and blockchain, an efficient, conservative, and self-sufficient energy industry.
Moreover, this paper analyzes and reviews two innovative application-specific use
cases of blockchain in the energy sector and one highly competent solidity code,
which could revolutionize the energy sector.
Keywords Blockchain ·Solidity ·Smart contracts ·Energy certificates
1 Introduction
Blockchain is a distributed ledger technology which is handled by peers on a peer-
to-peer network. It is an accounting system where a network of computers is used
K. M. Devi (B)
Department of Mathematics, Vignan’s Institute of Information Technology (A), Visakhapatnam,
AP, India
e-mail: mruduladevisai@gmail.com
D. S. Sai · N. T. Rao · K. Swathi
Department of Computer Science and Engineering, Vignan’s Institute of Information Technology
(A), Visakhapatnam, AP, India
S. Voddi
Department of Computer Science and Engineering, Prasad V. Potluri Siddhartha Institute of
Technology, Vijayawada, Andhra Pradesh, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_28
263
264 K. M. Devi et al.
Fig. 1 Blockchain representation [2]
to distribute the ledger. Hence fundamentally, blockchain technology is a record-
keeping tool [1]. Hierarchy is out of bound in blockchain. The network is handled
without it. Decentralization is the main virtue of using this technology. Hence, there
are not centralized storages or administrators. Blockchain w.r.t to trustworthy sources
could be comprehended as a series of blocks that could be incremented infinitely and
are connect via crypto-hashing methodologies. The blocks could be considered as
records too, and their timestamp (created time), data transactions, and the hash to the
preceding block are stored. Blockchain is designed to be resistant to any alterations
made to the data. The main agenda of it being a distributed ledger is its effective way
of withholding the transactions between parties, which are immutable in nature [2].
Alongside immutable records, numerous blockchains include discrete pioneering
contrivances for agents that sustain the integrity of network infrastructure, mainly
its data, thereby ensuring an adequately decentralized consensus. Three objectives
are contending for blockchain which are fast, low-cost, and decentralized, which is
described in Fig. 1. These three factors can give rise to a competent technology with
multi-folded use cases and many benefactor products.
A peer-to-peer network generally obeys a set of instructions or protocols for adding
new nodes into the chain, validating them and also for their implicit communication.
These all factor it to be a distributed ledger since the control is decentralized. The
moment a block is added into the chain, and modification of data like alteration and
removal is practically impossible, as shown in Fig. 1. To modify the data, the majority
of the network is required to give their consent, and here, the consensus algorithm
comes into play. Satoshi Nakamoto generally regarded as the father of blockchain
has brought the technology into existence in the year 2008 and on October 31st [3].
1.1 Applications
Blockchain has a various number of beneficial applications covering almost all
sectors. Table 1 will illustrate them in the most precise manner.
Blockchain and Its Idiosyncratic Effects on Energy Consumption … 265
Table 1 Applications of blockchain
Industry Use cases Start-ups
Energy, utility, and mining Smart utility metering system
Decentralized energy data
platforms
Bankymoon AutoGrid
Entertainment and media Control of ownership rights of
digital media
Disintermediation of record labels
Ascribe
Mycelia
Financial services International P2P transactions
Anti-money laundering
Bitcoin Confirm
Healthcare Storage of healthcare records
Population health and clinical
studies
HealthNautica Tierion
Insurance Peer-to-peer flight insurance
policies
Micro-insurance
InsurETH Stratumn
Freight transportation and logistics Trade documentation
Supply chain transparency
Wave Provenance
Hospitality Loyalty programs Loyyal
2 Literature Review
This paper consists of the use cases tagged with deep research work on the case
studies, and how it has affected the environment and energy systems in those areas.
This paper’s core objective is to provide a unique code that could help us develop
a smart contract to design the exchange procedure of energies within micro-grids.
The smart contracts can also help us add nodes securely into the blockchain in an
encrypted way. The consensus algorithm for validating new nodes into the chain is
also coded.
Considering blockchain to be the latest technology and considering its applications
in energy conservation to be idiosyncratic, there are very few existing systems that
have worked toward those regions. The papers: blockchain technology in the energy
sector: a systematic review of challenges and opportunities, blockchain for energy
utilities and how blockchain can be used for creating a market for energy savings
certificates and have provided an insight into the applications of energy certificates
and also blockchain can help the energy sector [4]. The citations of the research
papers given in the reference section have more detailed insight into blockchain’s
applications in trading, energy, or other products. They have been a driving force that
enabled me to code my smart contract explicitly for energy trading and to do further
detailed research on the case studies mentioned mainly by the author Gunther and
Andoni.
Though there are a few existing works on the blockchain with the energy sector,
this paper has detailed research on the case studies with reference to many local
journals, articles, and news. The Brooklyn micro-grid, though mentioned in many
266 K. M. Devi et al.
papers, has no papers that emphasize the amount of greener energy produced directly
or indirectly due to the inculcation of the blockchain technology into their micro-
grids. The concept of energy certificates though discussed in a reasonable amount
of research papers by Churong and by the article written by Business Stanford, here
we concentrate on the code which gave rise to it and that case study can be tagged
alongside a use case of an energy utility using blockchain technology. The process
of issuing energy certificates is both illustrated as a flowchart and also as a code for
a better understanding of it.
The core of this paper is the smart contract for validating the nodes to be added into
the chain. The nodes can be of any of the stake-holders like prosumers or generators of
the energy unit. Moreover, the smart contract also has the code to make the exchange
of energy functional. The exchange can be either via a medium of currency like any
cryptocurrency, or it can be a barter system that consists of the energy units as an
exchange medium.
2.1 Study and Review of the Factors/activities Involved
in the Energy Industry
Defining the factors/activities involved in the energy industry, use cases for energy
utilities via blockchain are vast. Understanding each use case/factor specified in
the energy sector is vital since it provides us with an idea to inculcate them as
blockchain’s applications in that field. They can be factorized via a few factors like
access, participation, domain, and model. Table 1 explains all the data tagging those
factors in the most comprehendible way:
Application [blockchain] specific factors and applications of blockchain in many
factors/use cases, mainly in the energy sector. However, there are specific wide-range
blockchain applications that could have a huge impression on the energy sector. This
paper identifies two such use cases, which are again tagged with already successful
case studies, which share some similarities with the use cases described below.
Use case 1: Induction of cryptocurrencies into implicit energy trading within a
micro-grid: A micro-grid is a closed energy eco-system where the transfer of energy
is done internally. It is generally used to manage energy efficiency by providing a
secure supply and mainly providing backup power in vital scenarios when there are
power outages. In such scenarios, we can ensure the possibility of benefiting the
citizens/people who produce locally by allowing them to sell their energy directly
to the people who are in need. Thus, by removing any 3rd party interference, we
can create a self-sufficient society that could be termed “smart” due to the complete
involvement of pure conventional energy resources. The removal of third-party inter-
ferences in such cases could profit on both sides, i.e., the consumer and the producer
[5]. The prices can be inferred directly based upon the demand and supply in the
micro-grid. In such cases, blockchain secures the transactions of both the sides, i.e.,
the producer/supplier and the consumer.
Blockchain and Its Idiosyncratic Effects on Energy Consumption … 267
Incorporation of smart contracts in such use cases can affect the following:
Smart contracts can be written via solidity, which can ensure the sufficient balance
of the demand- supply chain. Moreover, it improves energy efficiency by actively
balancing the chain.
Transactions are secured via hashes, which are implicitly encrypted, and further,
it has an immutable ledger that can prove any tampering of the resources.
Instead of traditional payment and confrontation, cryptocurrencies can be a
medium of payment between the two parties. Given that it is legalized in many
parts of the world, it is one of the safest options for transactions.
Example Case Study: The Brooklyn Micro-grid: Brooklyn micro-grid is a
blockchain-based P2P energy trading platform. It is located in the Gowanus and Park
slope communities in Brooklyn, NYC. A partnership runs it between LO3 Energy,
Consents, Siemens, and Centrica [6]. It enables the prosumers to directly sell their
energy surplus to their neighbors by using Ethereum-based smart contracts and PBFT
consensus, implemented by Tender mint. Micro-grids minimize the amount of energy
lost through transmission; as an estimated 5% of electricity created in the US is lost in
transit, micro-grids provide an efficient alternative. As already specified, the micro-
grid aims to provide electricity amidst a power outage. Hence, in this specific project,
the electricity is directed to the much-needed places like hospitals and shelters from
the houses during outages.
Profits were way out of reach to the residents (citizens) even when they could
garner, harvest, and sell their over-achieved energy produced by their photovoltaic
panels (PV). As an alternative, their bills were downsized. This is because t hey are
under the control of the company. It infers to the helplessness of the citizens in the
area since they cannot provide power during a shortage or blackout though they can
generate their power. This is because their PVs were shut down. Considering the
issues as mentioned above, the upgraded Brooklyn micro-grid ensured the authority
of the power generated with the respective citizens (residents) and thereby abol-
ishing the involvement of any third-party companies. In 2015, US solar developers
contributed 7.3GW of electricity to the grid, up from just 1GW in 2010, and a quarter
of this came from rooftop PV. An estimated 15% growth (annually) is expected in
the years ranging from 2018 to 2022 in the micro-grid market size globally. By 2022,
an expected 30 billion dollars is to be reached by this market.
Use case 2: Smart decentralized exchanging and contracting among energy
stake-holders in the global market: The above use case could be highly successful
at the regional level. Nevertheless, what if the energy needs to be exchanged among
large industries with varying motives or among different states? That is where this
use case comes into play, as we have already seen in the previous case study that
solar (green) energy has seen a spike, and the producers have seen profits alongside
the consumers who are having an uninterrupted supply of energy. The same can be
done throughout the world. However, we need an open and regulated system.
Moreover, the systems should be immutable, automated, and transparent. Smart
contracts become handy in such situations. This use case the contracts are signed
by the producers (energy) with numerous other parties. The consumers (energy)
ensure the tasks of getting parties onboard via contracting. Simultaneously, all the
268 K. M. Devi et al.
key bodies or stake-older stake-holders like the suppliers, distributors, and regula-
tors must approve the smart contracts and sign them. Blockchain nodes are owned
and administered by the key bodies. Entitlement to contrivance the signed terms of
contracts is also given to them. We will introduce energy saving certificates along-
side blockchain to enable all the above requirements. The agenda of implementing
and sculpturing this smart contract is to device an efficient platform to enable the
trading and implementing of energy saving certificates and thereby removing all the
third-party involvements [1, 4].
The function sell () is used to initialize the contract. It acts as a constructor and
declares the functions and variables implicitly. It sets the Ms. Sender as the owner
too. The modifier function, only owner alongside the sell () function, is executed by
the smart contract owner. It thereby is used to authenticate the task of selling the
overachieved () energy by the owner [6].
3 Proposed System: Application to Enable Energy Trading
[Ethereum Code]
The heart of the paper is concentrated here. The following two pages contain brief
solidity algorithms and pieces of code that could enable the application to globally
enable energy trading.
4 Conclusion
This paper mainly aims to show all the effects. Blockchain could have on the energy
sector, and in ways, it could revolutionize it. This paper consists of 2 main use cases,
which could be incorporated into the energy sector to enhance individual profits,
greener energy, and less power consumption. The use cases tagged along with them
have a similar kind of approach, which can be enhanced and made global. Given the
crisis of global warming, pollution, and scarcity of fossil fuels, the above factors could
enhance our sustainability on this Earth much longer. The smart contracts are written
can be used globally to help any individual or company to create a node/transfer
energy via them or use cryptocurrencies as a mode of commodity exchange. More-
over, the smart contract written is transparent, immutable, and decentralized. Hence,
it is the most viable source of trust. It can be finally concluded that blockchain can
help in a greener and self-sufficient society by making the changes as mentioned
above in the energy sector.
Blockchain and Its Idiosyncratic Effects on Energy Consumption … 269
References
1. Vasilevsky NA, Brush MH, Paddock H, Ponting L, Tripathy SJ, LaRocca GM et al (2013)
On the reproducibility of science: unique identification of research resources in the biomedical
literature. PeerJ 1:e148. https://doi.org/10.7717/peerj.148
2. Treiblmaier H (2019) Toward more rigorous blockchain research: recommendations for writing
blockchain case studies. Front. Blockchain 2:1–15. https://doi.org/10.3389/fbloc.2019.00003
3. Bhattacharyya D, Dinesh Reddy B, Kumari NMJ, Rao NT (2021) Comprehensive analysis on
comparison of machine learning and deep learning applications on cardiac arrest. J Med Pharm
Allied Sci 10(4):3125–3131. https://doi.org/10.22270/jmpas.V10I4.1395
4. Bhattacharyya D, Doppala BP, Thirupathi Rao N (2020) Prediction and forecasting of persistent
kidney problems using machine learning algorithms. Int J Curr Res Rev 12(20):134–139. https://
doi.org/10.31782/IJCRR.2020.122031
5. Scott B (2016) How can cryptocurrency and blockchain technology play a role in building
social and solidarity finance?. In: UNRISD working paper, No. 2016–1. Geneva, United Nations
Research Institute for Social Development (UNRISD)). Available online at: https://www.eco
nstor.eu/bitstream/10419/148750/1/861287290.pdf
6. Swanson T (2015) Consensus-as-a-service: a brief report on the emergence of permissioned,
distributed ledger systems. Technical report. Available online at: https://bit.ly/32q2bkt
Smart Hydroponics System for Soilless
Farming Based on Internet of Things
G. V. Danush Ranganath, R. Hari Sri Rameasvar, and A. Karthikeyan
Abstract Hydroponics agriculture is a soilless development strategy where the plant
is developed with the assistance of nutrient-rich supplements and water alone, and
consequently, it gives an answer for the developing shortage of horticultural land.
The point of this work is to plan and build an indoor programmed vertical hydroponic
framework that does not rely upon the external environment. The planned framework
is able to develop normal kinds of harvests that can be utilized as a food source inside
homes without the need of huge space. This process of cultivation is widely gaining
favor in countries where there are less arable lands by which the land required for
producing food crops keeps on decreasing. Specifically, land locked countries with
less sources of freshwater, deserts, and urbanized countries with larger developed
spaces than open lands. In this paper, a hydroponics cultivating stage is planned
and grown purposefully for the development of green feed expected for domestic
household purposes. The proposed work is to automate the process of hydroponics
through IoT implementation through which the cultivation can be monitored remotely
and also start correcting the environment to maintain stable growth. The aim of this
paper is the hope to develop a novel farming method that may well be the future of
farming.
Keywords Modern agriculture ·Hydroponics ·Aquaculture ·Nutrient film
technique ·Internet of things ·Automation ·ThingSpeak ·Remote sensors
G. V. Danush Ranganath (B) · R. Hari Sri Rameasvar · A. Karthikeyan
School of Electrical Engineering, Vellore Institute of Technology, Vellore 632014, India
e-mail: danushranganathg.v2018@vitstudent.ac.in
R. Hari Sri Rameasvar
e-mail: harisri.rameasvarr2018@vitstudent.ac.in
A. Karthikeyan
e-mail: karthikeyan.arun@vit.ac.in
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_29
271
272 G. V. Danush Ranganath et al.
1 Introduction
The world has changed beyond recognition since the dawn of agriculture as an
organized practice of cultivation. The land has changed, the air has changed, and the
water has changed. The traditional practice is muddles with blockades. The water
has changed in the way that they are no longer clean and or available as widely as
it used to be. The air has changed to reflect the industrial practices that have shaped
the modern world which highlights the ever changing atmosphere, more often for
the worse than for good [1]. The land has changed to satisfy the needs of the modern
occupants of this world so that now they can be called developed. And all these things
together led to the modern problem which puts agriculture at risk.
With modern problems come modern solutions and one such solution is hydro-
ponics. Hydroponics is the process of growing plants in sand, gravel, or water, the
surfaces which typically do not come to mind when one is thinking of farming [2].
With the world moving forward with its rapid urbanization and such development
leading to a reduction in the availability of arable lands, indoor farming seems like
the way to go.
As one of the significant supporters of hydroponics cultivators on the planet, China
has the biggest hydroponics industry. Besides, its hydroponics cultivation surpassed
50 million tons, representing over 60% of the world’s hydroponics cultivation in 2018
[3]. Gulf nations rely upon imports to meet 90% of their food and water utilization
needs and practice hydroponics for everyday needs.
IoT gadgets and programming applications are incorporated to send and show
framework data online. The utilization of IoT-based aqua farming mechanization in
this study guarantees that the information obtained is more effective, the information
obtained is supposed to be proficient because it is not done physically, however,
supplanted with the job of IoT.
2 Background Study
2.1 Origin and History of Hydroponics
Hydroponics is a branch of farming under the art of horticulture having agricultural-
ists grow medicinal and food crops for the world population. The term hydroponics
originates from two Greek expressions hydro referring to water and ponos meaning
labor [4]. The implementation of hydroponics farming was first introduced by Dr.
William Drederick Gericke, California professor from Nebraska, USA during early
1937.
Smart Hydroponics System for Soilless Farming … 273
2.2 What is Hydroponics
Hydroponics is a method of cultivation using nutrient-rich solutions in aqueous
medium for growth of the crops requiring no aid of soil or any other solid substances
during irrigation and its growth. The base idea of hydroponics is to promote plant’s
growth by providing flawless nutritional conditions to ensure exquisite production
[5].
Hydroponics cultivation has aced the traditional way of cultivation in numerous
fields including maximum yield of crops using small area for cultivation than in
conventional farming [6]. The hydroponics system is also effective in conservation
of water as the drained water can be recycled for further cultivation. It has the utmost
advantage of reducing the number of physical laborers in traditional methods.
This modern farming technique requires a huge cost to set up at the initial stage,
costs involve the acquisition of an educated team and building the setup for the plant’s
growth [7]. Hydroponics is vulnerable to power outages as it depends completely
on electricity for continuous monitoring and automation purposes. The hydroponic
yields tackle the pests but tend to fall into water borne diseases [8].
3 Technical Specification
3.1 Methodology
This project is trying to simulate an automated system of hydroponics where the
environment and water logging are controlled automatically and monitored remotely,
essentially implementing automation in the modern agricultural practice [9]. To do
the same, an IoT system with embedded sensors to closely monitor the environmental
conditions is used in tandem with a platform to monitor the changes logged by all
the sensors in the system. This system is also connected to a third party application
to put out alerts in case of any mishaps or abnormal changes logged in the system
(Fig. 1).
3.2 Hydroponic System and Essential Components
The proposed hydroponic system is designed using the nutrient film technique
(NFT), being a popular and most successful technique in hydroponics. It is typi-
cally prescribed to use polyvinyl chloride (PVC) pipe structures that are strong and
quite easy to channel water and maintain the cycle of logging and draining for NFT
systems as shown in the below Fig. 2.
Nutrients Container. The nutrient container is utilized to store the dissolved
nutrients in water that is provided to the NFT system. As a closed hydroponics
274 G. V. Danush Ranganath et al.
Fig. 1 Block diagram of the proposed system
Fig. 2 NFT hydroponic
system
system aims to recycle and reuse the excess water, the logging pipes are designed
such that the solution returns to it [10]. The material of the container used is plastic
that is ideally suggested. The number of crops is important for container setup and
is calculated using,
Number of plants(P) = no.of pipes(p) ∗ no. of shelves(s) ∗ no. of holes(h)(1)
Hence, as in Eq. (1), the number of plants that are grown here are 2 (p) *1 (s) *
3 (h) = 6 (P).
Wat e r Pum p. The water pumps and nutrient pumps are essential for optimum
calculation in the system. As this system is an entry level hydroponics system, a
12 V R385 diaphragm motor with specification of 1L/min of liquid flow is used. As
the measured value of the amount of water in the water logging pipe is around 1.5L,
Smart Hydroponics System for Soilless Farming … 275
R385 water pump seems to be a decent choice. The same pump is used to supply the
nutrient solution and pH solutions.
Nutrients and pH Solutions. The system is equipped with a Atlas pH sensor kit
with that monitors the pH levels in the solution for optimal growth of the plants. It is
suggested that a pH range between 4.5 and 7.5 is ideal for any hydroponic system. The
system’s solution i s combined with micromix hydroponics nutrient solution along
with radongrow pH up and pH down control solutions. The pH sensor is interfaced
to the Mega board through the analog pin A5, and its output is monitored through
the Arduino IDE and ThingSpeak cloud using the below conversion formula,
pH value(pH) ={[sensor voltage reading(V ) ∗ 5.0]/1024.0}/6(2)
pH value(calibrated) = pH value(pH) ∗ 3.5(3)
Real-time Clock Circuit. A DS3231 real-time clock module (RTC) based on
inter-integrated circuit (I2C) protocol was included to monitor real-time activities
in the system. The implementation of the module was programmed using a STMi-
croelectronics STM32F407 discovery board based on I2C protocol for real time and
date activities in the system and displayed using a 16 × 2 LCD display. The RTC
module is built with a coin cell battery to tackle power outages and keep a track on
the system.
Wat e r Flow P ath. The water flow path starting from the main water container is
through PVC pipes. A YF-S201 water flow sensor is placed in the path to have a track
and maintain the quantity of water flowing into the pipes. Water flow sensor used in
the system is connected to the analog pin A8 of the microcontroller having provided
required voltage to turn on by the microcontroller. The frequency of the signal is
calculated to calibrate the flow rate and amount of solution through the sensor. The
following mathematical expressions are used to compute flow rate and total volume
of the solution:
Sensor Frequency(Hz) = 7.5 ∗ Q(L/min), Q is flow rate in(L/min)(4)
Flow Rate(L/h) = (Sensor frequency ∗ 60 min)/7.5(5)
Liters = Q ∗ time elapsed(s)/60(s/min)(6)
Liters =[Frequency(Pulses/s)/7.5]∗ time elapsed(s)/60 (7)
Liters = Pulses/(7.5 ∗ 60)(8)
Actuators Control System. The proposed design uses a 8 channel 5 V relay
module that is connected with an Arduino Mega 2560 microcontroller to manage the
276 G. V. Danush Ranganath et al.
Fig. 3 Flow diagram of proposed system
water, pH, and nutrient pumps. The relay board is actuated based on the algorithm
with the readings of pH sensor, water level sensor float switch, DHT11 temperature
sensor, DS18B20 waterproof temperature probe, and BH1750 light intensity sensor
module. The water pumps come in contact with the power supply if a high pulse from
the microcontroller is triggered according to the algorithm implemented in Fig. 3.
Internet of Things (IoT) Platform. The Internet of things (IoT) is an evolving
technology and has contributed ideally in every field throughout the world. The IoT
feature in the proposed system brings in remote monitoring advantages for the user.
The real-time sensor readings including sensors such as temperature and humidity
sensor, pH, water level, and light intensity are sent to the server. The IoT feature is
implemented using the ESP8266 Wi-Fi module that pushes data to the ThingSpeak
cloud using the MQ telemetry transport (MQTT) protocol.
Figure 4 is the schematic diagram of the whole electronics circuit implemented
for real-time monitoring and automation using Fritzing software tool.
4 Project Implementation and Results
4.1 Hardware Setup
In this paper, the efforts have been taken to implement an actual hardware setup of the
proposed hydroponics system as in Fig. 5. The results of the implemented system
have been successfully verified with similar works on hydroponics. The designed
Smart Hydroponics System for Soilless Farming … 277
Fig. 4 Circuit diagram of the proposed system
system has been harvested with mint, producing maximum yield. The process of
growing crops in hydroponics systems is slightly longer than the traditional method
but having replaced it with the quality of food crops grown is an advantage.
4.2 ThingSpeak Dashboard
The proposed model of NFT-based hydroponics system was designed, developed,
and tested using various real-time sensor parameters including potential hydrogen
(pH), temperature, humidity, water temperature, water flow rate, and water float
level switch were continuously monitored and updated over the ThingSpeak cloud
for further analysis and remote monitoring. The following Fig. 6 includes the results
from the ThingSpeak Web interface cloud platform.
278 G. V. Danush Ranganath et al.
Fig. 5 Hardware setup of the system
Fig. 6 Sample graphs of ThingSpeak channel: a temperature, b humidity, c water temperature, d
pH level, e water flow, f water level
Smart Hydroponics System for Soilless Farming … 279
Fig. 7 Fresh mint crop
cultivation
4.3 Mint Cultivation
The developed hydroponics system was initially used to cultivate mint as it is an ideal
crop to grow in hydroponics. The parameters of the system are different for every
crop that is harvested in the system. The following Fig. 7 shows the outcome based
on the standard values from the developed hydroponics system with fresh yields.
5 Conclusion and Future Work
The ultimate aim of this project is to understand the working of hydroponics
using NFT and automate the system through IoT. The goal has been achieved by
constructing the system from the ground with references to existing models and
creating an innovative automated system. Through this project, we know the optimum
growing conditions for various plants, the right way to nourish a plant using only
water.
Continuing on the prototype, we can improve on the areas of scale, tempera-
ture control, and power consumption. Potentially, this could be addressed by imple-
menting renewable power resources, such as solar panels, instead of traditional DC
power outlets. The scale of the system can be developed into multiple shelves from
a simple single level structure for vertical hydroponics systems.
280 G. V. Danush Ranganath et al.
References
1. Muralimohan G, Arjun SV, Sakthivel G (2021) Design and development of IoT based hydro-
ponic farming setup for production of green fodder. Nat Volatiles Essent Oils 8(4):4325–4340
2. Chowdhury MEH, Khandakar A, Ahmed S, Al Khuzaei F, Hamdalla J, Haque F, Reaz MBI,
Shafei AA, Emadi NA (2020) Design, construction and testing of IoT based automated indoor
vertical hydroponics farming test-bed in Qatar. MDPI Sens 20:5637
3. Ullah A, Aktar S, Sutar N, Kabir R, Hossain A (2019) Cost effective smart hydroponic
monitoring and controlling system using IoT. Intell Control Autom 10:142–154
4. Khan S, Purohit A, Vadsari N (2021) Hydroponics: current and future state of the art in farming.
J Plant Nutr 44(10):1515–1538
5. Sharma N, Acharya S, Kumar K, Singh N (2019) Hydroponics as an advanced technique for
vegetable production: an overview. J Soil Water Conserv 17(4):364–371
6. Sambo P, Nicoletto C, Giro A, Pii Y, Valentinuzzi F, Mimmo T, Lugli P, Orzes G, Mazzetto
F, Astolfi S, Terzano R, Cesco S (2019) Hydroponic solutions for soilless production systems:
issues and opportunities in a smart agriculture perspective. Frontiers Planet Sci 10:923
7. Prasetia Y, Putrada AG, Rakhmatsyah A (2021) Evaluation of IoT-based grow light automation
on hydroponic plant growth. Jurnal Ilmiah Teknik Elektro Komputer dan Informatika (JITEKI)
7(2):314–325
8. Wei Y, Li W, An D, Li D, Jiao Y, Wei Q (2019) Equipment and intelligent control system in
aquaponics: a review. IEEE Access 7:169306–169326
9. Rajkumar R, Dharmaraj R (2018) A novel approach for smart hydroponic farming using IoT.
Int J Eng Res Comput Sci Eng (IJERCSE) 5(5)
10. Hadinata A, Mashoedah (2021) Internet of things-based hydroponic: literature review. IOP J
Phys: Conf Series 2111:012014
Solution Approach for Detection of Stock
Price Manipulation by Market Operators
Yogesh Kakde , Ganesh Chavan , Basant Sah , and Apoorva Sen
Abstract Nowadays, many so-called stock analyst people are now sending tips
via SMS, e-mails, and social media giving targets to s tocks that are of very bad
quality. In market, these are called pump and dump schemes, where “operators” or
“manipulators” increase the price of a stock by various strategies. The increase in
price attracts retail investor to purchase that stock. When the stock price crosses the
required targets, set by the manipulators who sell it out and public is left holding stock
whose price gets decreased suddenly. In this paper, we present a solution approach
which can be implemented to detect such manipulation of stock price and avoid such
malicious activity. In our solution approach, we suggest ideas, criteria which may
be used to build a model based on data analytics, machine learning which can return
us the list of stocks that is expected to be manipulated by the operator. This paper
also proposes the data analytic models, machine learning models which may be used
while implementing the solution approach suggested in this paper.
Keywords Data analytics ·Machine learning ·Stock market ·Stock price
manipulation ·Pumps and dumps
1 Introduction
Sometimes, there is a sharp pump or dump in the price of specific stocks. Although it
might happen as there may be chances that they are driven by operators or manipula-
tors. In other terms, the prices of those specific stocks have been moved or influenced
by the stock market operators. Let us take an example, the share price of Urja Global
Y. Kakde (B) · G. Chavan · B. Sah
KL University, Guntur, AP, India
e-mail: ykakde@gmail.com
B. Sah
e-mail: basantbitmtech2008@kluniversity.in
A. Sen
Medi-Caps University, Indore, MP, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_30
281
282 Y. Kakde et al.
stocks, it increased from around Re. 1 in Oct- 17 to Rs.11.43 in 2018 Jan and was
down to Rs. 2 by Oct—2018. It again started to rise in November 2018 but has been
on a long decline since then to Rs1.75 as of 24 Dec-2019. This manipulation was
done by operators.
Let us first understand who market operators are. Several brokers, speculators,
different types of firms, or sometimes even company persons may work together as
a syndicate to move a stock price for their personal agenda or profits. This syndicate,
also referred to as the stock market operators, work together to rapidly move stock
prices, creating a frenzy among the investors.
This is mostly done to get benefitted by the sharp and large price difference created
in the short period, so that higher profits may be generated. They target small and
mid- cap stocks as it is easier to manipulate and influence them.
1.1 Strategies to Influence Investors
Decision Order Book Manipulations
Different buying and selling orders for a specific stock are listed in order book. It
reflects the no of stocks, which people are willing to buy and sell at various prices.
Soft copy of order books is available and accessible to traders. The retail investors
can access only 5 data points, whereas the stock market operators may access much
more than that (Fig. 1).
Fig. 1 Number of market manipulation and price rigging cases taken up for investigation (Image
source: statistica.com)
Solution Approach for Detection of Stock Price … 283
Intraday Trading Tricks
Buying and selling of stocks within a single trading day are called intraday trading.
It is done to earn profits by getting advantage of the price movements. However, the
orders hence squared off by the end of the day, and hence, the name intraday trading
comes into the light.
1.2 How Does Manipulation Work?
Manipulation in stock prices can be done by various means. Decline in the price of
share can be fulfilled by placing a large quantity of small orders at a price that is
lower than current market price.
Investors feel that something wrong is going in the company. A negative sentiment
pushes investors to sell the stock that is why the price of the stock gets even lower.
Another way to increase the value of a share is by placing an equal number of buy
and sell orders for the same stock, but by different stock brokers. There are various
techniques of market manipulation.
The major techniques of market manipulation are:
Pump and Dump. It is used frequently in order to change the price of stock artifi-
cially. The manipulator now sells out the stocks at a greater price, and remaining are
left with an overestimated security which later goes down.
Poop and Scoop. It is not as frequently used as the pump and dump. In this technique,
the price of the stock of a medium or large-cap company is decreased artificially.
After that the manipulator buys the lower valued stocks which returns a profit.
2 Literature Survey
The discussion regarding internal attack is most important for the objective. Internal
attacks aim t o make profits by manipulation of trading processes, e.g., spoofing, quote
stuffing, layering, and others, which are the specific focus of this paper. Different
types of proprietary fraudulent activity detectors are deployed by stock exchanges to
analyze the time series data of trader’s activities or the activity of a particular stock
to flag potentially malicious transactions while human analysts probe the flagged
transactions further. [1] In study [2], we examine the phenomenon of stock touting
during pump and dump campaigns, in which deceivers advertise stocks to profit from
an increased price level.
Siering [3] Quantification and detection of pump and dump schemes that are coor-
dinated through Telegram chats and executed on Binance—one of the most popular
cryptocurrency exchanges. We detail how pumps are organized on Telegram and
284 Y. Kakde et al.
quantify the properties of 149 confirmed events with respect to market capitalization,
trading volume, price impact, and profitability.
We find that stocks with high levels of information asymmetry [4] and mid to
low levels of liquidity are most likely to be manipulated. A significant proportion of
manipulation occurs on month/quarter-end days.
Kakde et al. [5] This research motivates for implementation strategy design. The
main contribution of this research work is the design of a novel RNN-based ensemble
learning.
(RNN-EL) framework that combines trade-based features derived from trading
records and characteristic features of the list companies to effectively detect stock
price manipulation activities.
Insider trading samples occurred from year 2007 to 2017, and corresponding non-
insider trading samples are collected. Next, the proposed method is trained by the
GBDT, and initial parameters of the GBDT are optimized by the DE [6]. Finally,
out-of-samples are classified by the trained GBDT–DE model, and its performances
are evaluated.
3 Proposed Methodology
We propose a solution model which we may be called as an engine to detect stock
which is being manipulated. This engine detects manipulated stock at two levels. At
each level, we perform data collection and apply some statistical analysis.
The flow of our solution approach is (Fig. 2).
Level 1 Solution Approach. At this level, we may consider some features of the
stock. By applying statistical analysis, list out the possible stocks that have been/may
have been manipulated.
Level 2 Solution Approach. At this level, we consider only those stock which
have been filtered out after level 1 operations. At this level, we try to identify whether
the contents on the Web have been increased or not.
3.1 Detailed Discussion
Level 1. At this level, we perform only basic analysis. This analysis may include
features required for fundamental analysis and technical analysis [7]. For data collec-
tion, we only need to collect from only one source possibly. We can collect some
stats from the graph of stock price. These records of any stock may be given as input
to the data analytic model or machine learning model, e.g., decision tree, SVM, etc.,
which in return may give us an unpredicted pump or dump in the price of the stock.
At this level, we may collect the followings:
Solution Approach for Detection of Stock Price … 285
Fig. 2 Flow of solution
approach
•Sudden changes (hike) in the stock price crossing some threshold even if the
fundamentals of the company have not been changed.
•Order books can be accessed for the detection of stock price manipulation.
•Detection of bulk buying by any single investor.
•Sudden hike in demand for any stock.
•We check whether there are enough shreds of evidence for the stock price hike. The
evidence we may consider here is promoted holdings, funding received inflation,
market sentiment, etc.
Level 2. Once we figure out the possible list of stock which has been/may be
manipulated, we proceed for level 2 analysis. This level of analysis is computationally
more complex than level 1. In level 1, we may have only one or very few sources of
data for data gathering, but here in level 2, we need to access the Web.
•At this level, we try to collect the followings:
286 Y. Kakde et al.
Rapid growth in the keywords from
Social media contents: Access to open broadcasted message in the group
New articles: Small level of news media organizations
•Video contents on YouTube or other sites.
•Detection of buying stocks by the many numbers of investors without enough
change in the company’s fundamentals. (More complex than point II. discussed
in level 1)
•Data from the last 15 days will be analyzed.
3.2 Proposed Implementation Methods
Level 1. At this level, the input will be all possible stock which could be manipulated.
We can choose the basic parameters or some more parameters. After choosing the
required parameter, simply use any statistical tools to process the dataset. We suggest
using any one machine learning model from logistic regression, SVM, and decision
tree.
Level 2. At this level, we need to collect data from the Web, for that our search
engine uses a Web crawler [8]. A Web crawler is an Internet bot that systematically
browses the World Wide Web and that is typically operated by search engines for the
purpose of Web indexing. The Web crawler indexes the Web contents that contain a
specific keyword. The keyword could be anything which is related to the stock which
we are considering manipulated [9].
Crawler will collect the counts, URLs address, timestamps, etc., where the specific
keyword is found and will maintain a record for the required time span which could
be anything, e.g., 15, 30, 60 days, etc., [10] suggest 90 days a good time span to be
considered [11].
After data collection, we propose to implement advance data analytical model
which may be neural network model [12] or any model which can process multi-
dimensional data. The search engine will run continuously to collect data from the
Web. After a fixed time span, the analytics will be applied periodically on the data
collected [13].
4 Summary and Conclusion
This paper is focused on availing the solution method which could be used directly
and can be implemented using any data analytical tool or can be employed with a
machine learning model. Stock manipulation is very common nowadays in the stock
market, and there is no or very few solutions implemented to make investors aware of
this prior. This solution may be used by stock exchanges, brokers, and periodically,
they can warn their investors, especially retail investors to be aware of such malicious
activity. This paper provides solution at two levels, and this gives the flexibility to
Solution Approach for Detection of Stock Price … 287
choose to have a solutions implemented for level 1 only as the cost of implementing
and maintaining the level 2 solution is higher.
5 Future Work and Limitations
The methods discussed in this paper are enough to detect the stock(s) which have
been manipulated. In future, we can add some steps, not only for detection but also
for avoidance of the stock price manipulation. To avoid stock price manipulation,
we may come across with some good outputs like a list of investors who purchased
a particular stock in bulk, can take legal actions against this, we can identify the
media organizations, bloggers, and video bloggers who are involved in spreading
fake/bulk information. Stock brokers can also add constraints while buying stock in
bulk. We can also consider some more parameters for more accuracy. These could
be changed in CRR, Repo rate, and reverse r epo rate by Central Bank (e.g., RBI in
India), pandemic situations, and war conditions, because changes in these parameters
may also affect the stock price. The inclusion of this parameter will strengthen our
proposed engine.
As discussed above, we may collect a large amount of data from various sources
and for a predefined period. And then, we may apply analytics to that data. Collecting
more data for the long run and applying analytics may not be suitable for detecting
manipulation by operators in stock in case of intraday trading. This engine is limited
to finding operator manipulation overtime. So as far as intraday trading is concerned,
price manipulation may not be detected by this.
Bibliography
1. Kalra B et al (2021) Cooperative monitoring of malicious activity in stock exchanges. Pacific-
Asia conference on knowledge discovery and data mining. Springer, Cham
2. Nti IK, Adekoya AF, Weyori BA (2020) A systematic review of fundamental and technical
analysis of stock market predictions. Artif Intell Rev 53:3007–3057
3. Siering M (2019) The economics of stock touting during Internet-based pump and dump
campaigns. Inf Syst J 29(2):456–483
4. Victor F, Hagemann T (2019) Cryptocurrency pump and dump schemes: quantification and
detection. In: 2019 international conference on data mining workshops (ICDMW). IEEE
5. Kakde Y, Rawat M, Dangra J (2015) Implementation of efficient extraction of deep web by
applying structured queries and maintaining repository freshness. Int J Adv Res Comput Sci
Softw Eng 5:1266–1271
6. Comerton-Forde C, Putni ¸nš TJ (2014) Stock price manipulation: prevalence and determinants.
Rev Finance 18(1):23–66
7. Wang Q et al (2019) Enhancing intraday stock price manipulation detection by leveraging re-
current neural networks with ensemble learning. Neurocomputing 347:46–58
8. Kakde Y, Agrawal S (2018) Predicting survival on titanic by applying exploratory data analytics
and machine learning techniques. Int J Comput Appl 179(44):32–38. https://doi.org/10.5120/
ijca2018917094
288 Y. Kakde et al.
9. Heydon A, Najork M (1999) Mercator: a scalable, extensible web crawler. World Wide Web
2(4):219–229
10. He J, Xia H, Zhao Y (2021) Pump-and-Dump” through media tone? Institutional trading strate-
gies during corporate litigation. Institutional Trading Strategies during Corporate Litigation
(December 6, 2021)
11. Wang Z et al (2022) Design and implementation of security vulnerability sharing plat- form
based on web crawler. In: Proceedings of the 11th international conference on computer
engineering and networks. Springer, Singapore
12. Skillicorn D, Nam D (2022) Detecting stock market manipulation from online forums
13. Kakde Y, Bothe N, Paul A (2019) Real life implementation of object detection and classification
using deep learning and robotic arm. SSRN Electron J 2–9. https://doi.org/10.2139/ssrn.337
2199
Cancer Cell Detection and Classification
from Digital Whole Slide Image
Anil B. Gavade , Rajendra B. Nerli , Shridhar Ghagane ,
Priyanka A. Gavade, and Venkata Siva Prasad Bhagavatula
Abstract The World Health Organisation has identified cancer as one of the fore-
most causes of death globally which reports that nearly one in six deaths is due
to cancer. Hence, an early and correct diagnosis is required to assist doctors in
selecting the accurate and best treatment option for the patient. Pathological data
have huge tumour information that can be used to diagnose cancer. Digitizing patho-
logical data into images and its analysis using Deep learning applications will be
a significant contribution to clinical testing. Due to advancements in technology,
artificial intelligence (AI) and digital pathology can now be combined allowing for
image-based diagnosis. This study uses residual networks (ResNet-50) and convolu-
tional neural network (CNN), which is pre-trained on ImageNet dataset to train and
categories lung histopathology images into non-cancerous, lung adenocarcinoma,
and lung squamous cell carcinoma delivering an accuracy of 98.9%. Experimenta-
tion results show that the ResNet-50 model delivers finer classification results when
compared to state-of-the-art methods.
A. B. Gavade (B)
Department of E&C, KLS Gogte Institute of Technology, Belagavi, Karnataka, India
e-mail: anil.gavade@gmail.com
R. B. Nerli
Department of Urology, JN Medical College, KLE Academy of Higher Education and Research
(Deemed-to-Be-University), Belagavi, Karnataka, India
S. Ghagane
Department of Biotechnology, KAHER’s Dr. Prabhakar Kore Basic Science Research Center, V.
K. Institute of Dental Sciences Campus, Belagavi, Karnataka, India
P. A. Gavade
Department of Computer Science and Engineering, KLE Tech University Dr. M. S. Sheshgiri
College of Engineering and Technology, Belagavi, Karnataka, India
V. S. P. Bhagavatula
Medtronic, Hyderabad, India
e-mail: Venkata.siva.prasad.bhagavatula@medtronic.com
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8_31
289
290 A. B. Gavade et al.
Keywords Histopathological image ·Deep learning ·Convolutional neural
network ·Classification ·Artificial intelligence ·Residual networks ·Graphic
processing unit
1 Introduction
Cancer is a disease that has a very low survival rate, accounting for nearly 10 million
deaths in 2020. The most common in them were breast and lung. Due to the high
recurrence and death rates, the treatment is lengthy and costly. Cancer early stage
prognosis is not easy, due to lack of availability of diagnostic tools that are critical
in clinical cancer research. Accurate early cancer identification and prognosis are
crucial for improving the patient’s survival rate.
Pathology is a branch of medical science that studies and diagnoses disease by
analysing surgically removed tissues, organs, fluids and in some instances, even the
entire body. Pathology also encompasses the closely related scientific study of disease
processes, which examines the causes, mechanisms, and consequences of illness.
Under a microscope, the pattern of tissue samples is examined to help determine
whether a sample is malignant or not. This requires a lot of time investment and labour
work, resulting in reduced efficiency of workflow. Thus, digitising this process will
help in increased work efficiency and faster diagnosis. This can be achieved by digital
pathology, which is a dynamic and image-based platform that allows pathology data
generated from a digitised glass slide to be acquired, managed, and interpreted.
With the practice of whole slide imaging, glass slides can be converted into digital
slides that may be viewed, managed, and analysed on a computer. Development of
AI and machine learning leads to efficient and less expensive disease diagnosis,
prognosis, and prediction systems.
Deep learning extracts biomarkers from histology images directly and summarises
the research of cancer histology image analysis. They are designed to automate work-
flows. These algorithms can be used for the purpose of segmentation and classification
of whole slide images.
In digital pathology, first, we classify whole slide images as cancerous or non-
cancerous; further, we use segmentation to identify the size and location of the cancer,
all this is achieved through training the model on convolutional neural networks
(CNNs). CNN is a type of deep neural network that is commonly used to analyse
visual data and has a pre-trained learning model for image classification. The model
also contains nearly 23 million trainable parameters, indicating a deep architecture.
ResNet is one such CNN that can be used for classification of image-based dataset.
ResNet-50 is a 50 layers deep CNN that can load a pre-trained network model from
the I mageNet database that has been trained on over 1 lakh images. There are 48
convolution layers, one Maxpool layer, and one average pool layer in the model.
The ResNet-50 model has 5 stages. Each stage has a residual block. These residual
blocks have 3 layers, each with 1 × 1 and 3 × 3 convolutions. Each convolutional
layer is followed by a batch normalisation layer and a ReLU activation algorithm. In
Cancer Cell Detection and Classification from Digital Whole Slide Image 291
traditional neural networks, each layer feeds into the next, but in ResNet, each layer
feeds directly into the next layer and onto layers 2/3 hops away. These are known as
identity connections.
2 Contributions
The paper is implement for cancer detection and classification using digital whole
slide and deep neural network architecture, i.e. CNN models. From literature, it is
observed CNN which is considered as one of the best pre-trained model for large
data image classification applications, and we find ResNet-50 widely used in medical
image classification applications. The paper mainly divided into literature review,
implementation, performance comparisons analysis, results, and conclusion.
3 Related Works
Image segmentation is one of the important and challenging task in the area of
medical image processing. For nucleus segmentation,
Shyam Lal et al. [1] proposed an encoder–decoder style U-net model with an
attention-gating mechanism and a dimension wise pyramid pooling approach. The
model was evaluated on kidney and breast histopathology images that resulted in
a F1-score of 0.9294 and an average Jaccard index (AJI) of 0.8688 for publicly
available kidney dataset, and a F1-score of 0.8243 and an AJI of 0.7039 for breast
dataset.
Zitao Zengm et al. [2] presented a model which uses a multi-task learning tech-
nique to segment nuclei and cell contour simultaneously. This model delivered the
F1-score as 0.8278 and the dice score as 0.7844.
Amit Kumar et al. [3] proposed separable convolution pyramid pooling network
with an encoder–decoder. Evaluation was done on kidney and breast datasets which
resulted to give F1-scores as 0.9203 and 0.8168 and AJI as 0.8592 and 0.6998 for
kidney and breast datasets, respectively.
Shyam Lal et al. [4] presented an architecture having three blocks. A robust
residual block, bottleneck block and an attention decoder block. To extract high
level sematic maps robust residual block is proposed, along with attention block for
effective object localization this improved the architecture. During segmentation, the
model claims to be more precise in tackling shape variability and nuclei contacting
challenges.
Qasem et al. [5] used the ResNet-50 CNN pre-trained model on ImageNet to
categorise the dataset into benign and malignant category. This model was compared
with various CNN models, and it was found that the proposed method has the accuracy
of 99.10% outperforming the state-of-the art methods.
292 A. B. Gavade et al.
Pin Wang et al. [6] gave an architecture for automatic segmentation and classifica-
tion of breast histology images. The method uses wavelet transformation and multi-
scale region growing to detect the regions of interest and morphological operation
along with a CSS detection algorithm to separate the overlapped cells.
Anmol Verma et al. [7] proposed a model prediction and classification of breast
cancer histopathology images. For detection, the model was influenced by the
IRRCNN algorithm, and for classification, it was influenced by WSI-Net. The accu-
racies for detecting the cancer and classifying them were 95.25% and 80.43%,
respectively, which outperformed WSI-Net.
Muhammed Talo [8], presented deep learning ResNet-50 and DenseNet -161
models to classify histopathology images automatically, with accuracy of 97.89%
(Gray image and colour) and with dataset Kima Path 24 datasets as 24 classes with
classification accuracy of 98.87%.
Yangqin Feng et al. [9] implemented cell nuclei classification using Breast Cancer
histopathology images using stacked denoising autoencoder and compared with 8
different techniques and their classification accuracy of 98.27% result outperformed
with others.
Yun Jiang et al. [10] presented a small SE-ResNet module that combines a residual
module with a squeeze-and-excitation block. The model classifies histopathological
images of breast cancer into benign, malignant, and eight subgroups. For binary
classification, the achieved accuracy ranges between 98.87 and 99.34% and 90.66
and 93.81% for multi-class classification.
Krithiga and Geetha [11] published a review on detection, segmentation, and
classification of breast histopathology images. The study provides an overview of
tissue preparation, stained image analysis, preprocessing techniques, methods of
segmentation, methods of feature extraction, feature selection, and classification.
This work drew attention to several algorithms and methodologies, as well as listing
the performance of various models with various characteristics such as accuracy,
specificity, sensitivity, and F1-score.
Kourou et al. [12] present a review overview of cancer prognosis and predic-
tion using machine learning in this literature they have considered papers related
to different cancers like oral cancer, Brain Cancer, Colon cancer, Cervical cancer
addressed cancer detection and classification using ANN, SVM, Graph based SSL
algorithm.
Mesutetal. [13] proposed a model that includes an attention module, hypercolumn
technique and a residual block for improved cancer detection. When evaluated on
BreakHis dataset, this model achieved an accuracy of 98.80%.
Soulami et al. [14] used DDSM and INbreast mammographic database for breast
cancer automatic segmentation and classification of breast cancer, proposed an end-
to-end U net model, results are assessed with evaluation matrices such as IOC, AUC,
F1 and dice coefficient.
Ting- Wei Chiu et al. [15], addressed the position of lung nodule lung cancer and
carcinoma with U-Net and 2DU-Net segmentation architecture. Evaluation matrixes
used were Dice coefficient, accuracy, sensitivity and specificity comparisons carried
Cancer Cell Detection and Classification from Digital Whole Slide Image 293
with data without preprocessing (mono input positive) and with processing (mono
input negative).
Devvi Sarwinda et al. [16], implemented CNN ResNet model architecture ranging
from 18 layers to 152 layers. Colorectal gland image dataset tested on ResNet-18 and
50, different set of training and testing ratios are considered for results verifications,
classification results assessed with accuracy, sensitivity and specificity, it is observed
that, the higher number of layers too more time for computation, final inference was
with less number in ResNet architecture, it is possible to achieve good classification
accuracy with less time.
Brij Rokad and Nagarajan [17] demonstrated skin cancer detection and classifica-
tion using Deep Residual Network (ResNet) for International Skin Imaging Collabo-
ration (ISIC)-2017 challenge skin dataset (dermoscopic lesion images) around 2000
images (374 melanoma 254 Seboherric Keratosis and 1372 Nevus (Begnin)) achieved
classification accuracy of 77%.
Jiazhi Liang [18] implemented CNN ResNet-110 V1 for classification of CIFAR-
10 datasets, different training and testing combination used for experimentation and
it is observed to have the highest accuracy at 110 layers.
Yasi n Yar i et al. [19], developed an effective training-learning architecture that
consists of fully connected classifier and input layers combined with the ResNet-
50 and DensNet-121 model. Different magnifying images are employed to test the
proposed techniques with 8 other techniques, binary and multi labeled classification
algorithms tested on histology datasets.
Varsha Prakash and Smitha Vas [20], reviewed lung cancer using modalities like
X-ray and CT-Image, this paper overview with segmentation and nodule extraction,
nodule classification and emphasis on CNN data augmentation and nodule detection.
Sham lal et al. [21], demonstrated segmentation of nucleic cell from stained histo-
logical slide, implemented algorithm is compared with four different methods for
results performance assessment. Algorithm experimented with two different datasets
such as Stephan Wienert and liver tissues datasets from KMC Mangalore. Results
are compared with parameters such as precision, recall and F1 as quality metrics and
implementation is outstanding with other 4 algorithms.
Hao Dong et al. [22], used BRAT-2015 dataset to develop the U-net CNN algo-
rithm, which segments patient specific brain tumors without manual intervention
and this potentially enables objective lesion assessment for clinical tasks such as
diagnosis, treatment planning and monitoring.
Amitojdeep et al. [23], reviewed and addressed several diseases with different
modalities and classification of interested areas from radiological modalities. The
application of modalities includes brain MRI, X-ray, Cardic MRI, CT, mammography
and lung CT. Focus of review is on classification and segmentation architectures with
CNN and it derivatives models, SVM and Hybrid CNN.
Ves a l et a l . [24] provide a performance comparison between ResNet-50 and
Inception-V3 which have been pre-trained on ImageNet dataset and then trained
on BACH dataset. A transfer learning-based approach is been proposed, in which
ResNet-50 achieves 97.50% accuracy outperforming Inception-V3 with 91.25%
accuracy.
294 A. B. Gavade et al.
Nur Syahmi Ismail and Cheab Sovuthy [25], addressed breast cancer with dataset
IRMA, implementation comparisons done with three different techniques, VGG-16,
ResNet-50 and implementation carried out by Q. Zhang, aim is to classify as normal
or abnormal tumor, the results are assessed with precision, accuracy and recall as
quality matrices, it is observed VGG-16 outperformed other two algorithms.
Asmaa Hekal et al. [26], developed deep learning model for breast cancer detection
and classification using dataset CBIS-DDSM ROI dataset. The CNN model is refined
at last fully connected layer of the pre-trained model is substituted with a SVM
shallow classifier, this lead with improved classification of tumors.
Hao Zhang et al. [27] propose a ResNet model for detection of metastatic cancer,
and test time augmentation is employed in the model to make it more robust and to
improve detection accuracy.
Saber et al. [28] proposed a DL model for enhancing the classification results using
transfer learning on the Mias dataset. The VGG-16 model achieved the best accu-
racy of 98.96% compared to ResNet-50 Inception_V3, VGG-19, and Inception_V2
ResNet.
Ahmad et al. [29] present the use of transfer learning for classification of
breast cancer. The ResNet-50 model has achieved an 85% accuracy on image-wise
classification and 83.60% on patch-wise classification on BreakHis dataset.
Shallu Sharma and Rajesh [30], proposed comparison of hand crafted features
to conventional classifiers and transfer learning baseline CNN pre-trained model
for feature extraction with classification with CNN classifier. Breast cancer
histopathology datasets used for experimentation, the datasets used are in different
magnification (40X, 100X, 200X and 400X). It is observed that the pre-trained model
as feature extractor outperformed the hand crafted features to conventional classifiers
for different magnification images.
4 Dataset and Computing Machine Details
In the implementation, only the lung organ dataset [31] is used with a total of 3
classes and 15,000 images, patch dimension of 768 × 768 in colour JPEG format.
(1) Lung: Benign tissue (5000 samples), adenocarcinoma (5000 samples), and
squamous cell carcinoma (5000 samples)
(2) Colon: Benign tissue (5000 samples) and adenocarcinoma (5000 samples).
The dataset has 25,000 histopathological images which divided into five classes.
All the images are 768 × 768 pixels in size and saved as jpegs. The images were
augmented to 25,000 that included a total of 750 images of lung tissue and 500 images
of colon tissue. Algorithm implemented on Dell Precision Tower 5810 work station
with specification Xeon CPU, 512 GB SSD, 32 GB RAM, and 8 GB Quadro P4000
Nvidia GPU.
Cancer Cell Detection and Classification from Digital Whole Slide Image 295
Fig. 1 Whole slide image cancer classification
5 Implementation Details
The implementation block diagram is represented in Fig. 1, which involves datasets,
ResNet-50 CNN model and test image. The model is trained, tested, and vali-
dated with different percentage combination of dataset, and it is observed the
model performed efficiently with 70:15:15 ratio combination. Algorithm 1 explains
the procedure for cancer classification. Samples of digital histology images are
shown in Fig. 2, which consist non-cancerous tissue, malignant tissue of type lung
adenocarcinoma and malignant tissue of type lung squamous cell carcinoma.
Algorithm 1: Cancer cell detection and classification
(continued)
Fig. 2 Biopsy sample data a non-cancerous tissue, b malignant tissue of type lung adenocarcinoma,
c malignant tissue of type lung squamous cell carcinoma
296 A. B. Gavade et al.
(continued)
Input: Digital whole slide image
Output: Non-cancerous tissue/is malignant tissue of type lung
adenocarcinoma/malignant tissue of type lung squamous cell carcinoma
Step 1: Importing required libraries and loading lung cancer dataset
Step 2: Data pre-processing and data splitting with balanced test, train split
Step 3: Loading ResNet-50 model
Step 4: Defining final output layers as 3 same as number of classification folders
Step 5: Compiling the model with loss and optimising functions as categorical cross
entropy and Adam, respectively
Step 6: Executing the model
Step 7: Plotting confusion matrix
Step 8: Evaluating the model
Step 9: Prediction and classification of lung cancer into its subtypes
6 Experimental Results
The dataset has 25,000 histopathological images which divided into five classes. All
the images are 768 × 768 pixels in size and saved. The ResNet-50 is trained, tested,
and validated with 70:15:15 ratio combination performance analysis of accuracy and
loss which is shown in Fig. 3.
The data tested for five different pre-trained CNN models like VGG-16, Effi-
cientNetB0, EfficientNetB7, AlexNet and ResNet-50. The accuracy parameter is
considered as performance concluding parameter shown in Table 1, amongst these
tested model ResNet-50 outperformed, Table 1 shows the classification accuracy of
0.989, and this is the highest achieved classification accuracy.
Fig. 3 Performance analysis a accuracy, b loss
Cancer Cell Detection and Classification from Digital Whole Slide Image 297
Table 1 Comparative discussion accuracy metric
Model VGG-16 EfficientNetB0 EfficientNetB7 AlexNet ResNet-50
Accuracy 0.9763 0.970 0.9656 0.948 0.989
Table 2 Comparative discussion ResNet-50
Class Precision Recall Support F1-score
Lung adenocarcinoma 0.926 0.951 0.948 0.932
Lung benign 0.909 0.921 0.935 0.941
Lung squamous cell carcinoma 0.951 0.950 0.942 0.938
The data tested for five different pre-trained CNN models like VGG-16,
EfficientNetB0, EfficientNetB7, AlexNet, and ResNet-50.
The implementation is assessed with three different datasets, and classification
performance comparisons are made with reference to precision, recall, support and
F1-score parameters shown in the Table 2.
7 Results Overview
The implementation performance analysis presented in above tables, different deep
neural network—CNN model comparisons is shown in Table 1, and classification
accuracy is considered. From Table 2, it is clear that ResNet-50 outperforms other
models, trained on the same parameters achieving an accuracy of 98.9% and also
properly classifying the tissue into its cancer subtypes classified as non-cancerous or
cancerous tissue. From the plots Fig. 3, we can infer that there is a drastic accuracy
increase and proportional loss decrease till the first epoch, then a gradual increase in
accuracy and decrease in loss after the first epoch till fifth epoch.
8 Conclusion
This paper presents implementation of digital whole slide image cancer detection
and classification using deep neural network CNN pre-trained model ResNet-50. The
performance of CNN is much superior with high classification accuracy. In conclu-
sion from the results obtained, ResNet-50 outperforms compare to other models. As
a future scope, the algorithm could be implemented with U-net for segmentation and
classification accuracy still further improved. Also we can extract large number of
features using CNN, dimension reduction could be done using principal component
analysis (PCA) and optimised the code for higher processing, improved accuracy
with the help of graphic processing unit computing architecture.
298 A. B. Gavade et al.
References
1. Aatresh AA, Yatgiri RP, Chanchal AK, Kumar A, Ravi A, Das D, Raghavendra BS, Lal S, Kini
J (2021) Efficient deep learning architecture with dimension-wise pyramid pooling for nuclei
segmentation of histopathology images. Comput Med Imaging Graph 93:101975
2. Zeng Z, Xie W, Zhang Y, Yao L (2019) RIC-Unet: An improved neural network based on Unet
for nuclei segmentation in histology images. IEEE Access 7:21420–21428
3. Chanchal AK, Kumar A, Lal S, Kini J (2021) Efficient and robust deep learning architecture
for segmentation of kidney and breast histopathology images. Comput Electric Eng 92:107177
4. Lal S, Das D, Alabhya K, Kanfade A, Kumar A, Kini J (2021) NucleiSegNet: robust deep
learning architecture for the nuclei segmentation of liver cancer histopathology images. Comput
Biol Med 128:104075
5. Al-Haija QA, Adebanjo A (2020) Breast cancer diagnosis in histopathological images using
ResNet-50 convolutional neural network. In: 2020 IEEE international IOT, electronics and
mechatronics conference n(IEMTRONICS), IEEE, pp 1–7
6. Wang P, Xianling H, Li Y, Liu Q, Zhu X (2016) Automatic cell nuclei segmentation and
classification of breast cancer histopathology images. Signal Process 122:1–13
7. Verma A, Panda A, Chanchal AK, Lal S, Raghavendra BS (2021) Automatic deep
learning framework for breast cancer detection and classification from H&E stained breast
histopathology images. In: Data science. Springer, Singapore, pp 215–227
8. Talo M (2019) Automated classification of histopathology images using transfer learning. Artif
Intell Med 101:101743
9. Feng Y, Zhang L, Yi Z (2018) Breast cancer cell nuclei classification in histopathology images
using deep neural networks. Int J Comput Assist Radiol Surg 13(2):179–191
10. Jiang Y, Chen L, Zhang H, Xiao X (2019) Breast cancer histopathological image clas-
sification using convolutional neural networks with small SE-ResNet module. PLoS ONE
14(3):e0214587
11. Krithiga R, Geetha P (2021) Breast cancer detection, segmentation and classification
on histopathology images analysis: a systematic review. Archives Comput Methods Eng
28(4):2607–2619
12. Kourou K, Exarchos TP, Exarchos KP, Karamouzis MV, Fotiadis DI (2015) Machine learning
applications in cancer prognosis and prediction. Comput Struct Biotechnol J 13:8–17
13. To ˘gaçar M, Özkurt KB, Ergen B, Cömert Z (2020) BreastNet: a novel convolutional neural
network model through histopathological images for the diagnosis of breast cancer. Phys A:
Stat Mech Appl 545:123592
14. Soulami KB, Kaabouch N, Saidi MN, Tamtaoui A (2021) Breast cancer: one-stage automated
detection, segmentation, and classification of digital mammograms using UNet model based-
semantic segmentation. Biomed Signal Process Control 66:102481
15. Chiu T-W, Tsai Y-L, Shun-Feng S (2021) Automatic detect lung node with deep learning in
segmentation and imbalance data labeling. Sci Rep 11(1):1–10
16. Sarwinda D, Paradisa RH, Bustamam A, Anggia P (2021) Deep learning in image classification
using residual network (ResNet) variants for detection of colorectal cancer. Procedia Comput
Sci 179:423–431
17. Rokad B, Dr. Nagarajan (2019) Skin cancer recognition using deep residual network. arXiv
preprint arXiv:1905.08610
18. Liang J (2020) Image classification based on RESNET. J Phys: Conf Series 1634(1):012110.
IOP Publishing
19. Yari Y, Nguyen TV, Nguyen HT (2020) Deep learning applied for histological diagnosis of
breast cancer. IEEE Access 8:162432–162448
20. Prakash V, Vas PS (2020) Survey on lung cancer detection techniques. In: 2020 international
conference on computational performance evaluation (ComPE), IEEE, pp 800–803
21. Lal S, Desouza R, Maneesh M, Kanfade A, Kumar A, Perayil G, Alabhya K, Chanchal AK,
Kini J (2020) A robust method for nuclei segmentation of H&E stained histopathology images.
Cancer Cell Detection and Classification from Digital Whole Slide Image 299
In: 2020 7th international conference on signal processing and integrated networks (SPIN),
IEEE, pp 453–458
22. Dong H, Yang G, Liu F, Mo Y, Guo Y (2017) Automatic brain tumor detection and segmentation
using U-Net based fully convolutional networks. In: Annual conference on medical image
understanding and analysis. Springer, Cham, pp 506–517
23. Singh A, Sengupta S, Lakshminarayanan V (2020) Explainable deep learning models in medical
image analysis. J Imaging 6(6):52
24. Vesal S, Ravikumar N, Davari AA, Ellmann S, Maier A (2018) Classification of breast cancer
histology images using transfer learning. In: International conference image analysis and
recognition. Springer, Cham, pp 812–819
25. Ismail NS, Sovuthy C (2019) Breast cancer detection based on deep learning technique. In:
2019 international UNIMAS STEM 12th engineering conference (EnCon), IEEE, pp 89–92
26. Hekal AA, Elnakib A, Moustafa H-D (2021) Automated early breast cancer detection and
classification system. SIViP 15(7):1497–1505
27. Zheng Z, Zhang H, Li X, Liu S, Teng Y (2021) Resnet-based model for cancer detection.
In: 2021 IEEE international conference on consumer electronics and computer engineering
(ICCECE), IEEE, pp 325–328
28. Saber A, Sakr M, Abo-Seida OM, Keshk A, Chen H (2021) A novel deep-learning model for
automatic detection and classification of breast cancer using the transfer-learning technique.
IEEE Access 9:71194–71209
29. Ahmad HM, Ghuffar S, Khurshid K (2019) Classification of breast cancer histology images
using transfer learning. In: 2019 16th international bhurban conference on applied sciences
and technology (IBCAST). IEEE, pp 328–332
30. Sharma S, Mehra R (2020) Conventional machine learning and deep learning approach for
multi-classification of breast cancer histopathology images—a comparative insight. J Digit
Imaging 33(3):632–654
31. https://www.kaggle.com/datasets/andrewmvd/lung-and-colon-cancer-histopathological-
images Accessed on 30 May 2022
Author Index
A
Abdul Rahman, M., 215
Ali Baig Mohammad, 161
Aneel Kumar Minda, 81
Anil B. Gavade, 289
Apoorva Sen, 281
Asish Vardhan, K., 179, 251
B
Basanaboyana Vamsi Sai, 147
Basant Sah, 281
Bondo, Espoir M. M., 109
C
Chandana Seelam, 71
Charan Naga Santhu Jagadeesh Boddeda,
171
Chris Daniel Mantri, 171
D
Danush Ranganath, G. V., 271
Debdatta Bhattacharya, 179, 189
Debnath Bhattacharyya, 61, 117, 129, 139,
147, 153, 171, 197, 241, 251
Devi Satya Sri, V., 91
Devi Shanthisree Kumpatla, 171
Dhiren Dommeti, 205
Dinesh Reddy, B., 147, 171, 241
Doddala Jyo-theendra, 139
E
Eali Stephen Neal Joshua, 179, 189, 197,
241, 251
Eswar Abisheak Tadiparthi, 147
F
Fardeen,S.K., 51
G
Ganesh Chavan, 281
Geetha, G., 171
Gnana Jeevana, K., 129
Gonuguntla Krishna Mohan, 223
H
Hari Sri Rameasvar, R., 271
J
Jai Sai Chaitanya, K., 51
Jampani Satish Babu, 223
Jyothsna, M., 129
K
Kannuru Chandana, 139
Karthikeyan, A., 271
Karthik, S., 129
Kollana Bharat Kalyan, 147
Kopanathi Mouli, 41
© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Singapore Pte Ltd. 2023
K. A. Ogudo et al. (eds.), Smart Technologies in Data Science and Communication,
Lecture Notes in Networks and Systems 558,
https://doi.org/10.1007/978-981-19-6880-8
301
302 Author Index
M
Majji Prasanna Kumari, 147
Mamidi Sai Sri Venkata Spandhana, 153
Manoj Challapalli, 71
Marline Joys Kumari, N., 153
Mohammed Ismail, B., 161
Mohd. Abdul Muqeet, 161
Mrudula Devi, K., 263
Muteba, Arcel Kalenga, 101, 109
Muthumanickam, K., 1
Muzammil Parvez, M., 161
N
NagaMallik Raj, S., 61, 129, 139
Neeraja, S., 61
Nikhil, P., 139
Nirujogi Venkata Sai Sandeep, 153
Nithin Sai, N., 215
O
Ogudo, Kingsley, A., 101, 109
P
Palani Kumar, R., 1
Pandiaraja, P., 1, 15
Paturi Jyothsna, 153
Penki Ganesh, 41
Pisupati Krishna Teja, 29
Prabhakar, G., 51
Praveena, N., 223
Priyanka A. Gavade, 289
Puppala Ramya, 29, 41, 51, 215
Pyla Lohit, 139
Q
Quazi Mateenuddin Hameeduddin, 161
R
Rajendra B. Nerli, 289
Rao, N. Thirupathi, 61, 129, 139, 147, 153,
171, 179, 189, 197, 241, 251, 263
Rayi Jayasri, 153
S
Sai Mokshith, M., 215
Sangeeta Parshionikar, 117
Satyanarayana, K. V., 189, 197
Satyanarayana, M., 197
Shaik Qadeer, 161
Shridhar Ghagane, 289
Siva Rama Krishna Nallapati, 205
Srikanth Vemuru, 91
Srinivas, P. V. V. S., 205
Srinu, P., 129
Sumathi, K., 15
Surya Sai, D., 263
Swathi, K., 153, 257, 263
Swathi Voddi, 263
T
Tadepally Hrushikesh, 29
Tejaswi Talluru, 71
Tummala Haswanth Chowdary, 29
V
Vamsi Krishna, G., 257
Varaha Sai Adireddi, 171
Venkata Naresh Mandhala, 205
Venkata Siva Prasad Bhagavatula, 289
Vithya Ganesan, 71, 81
Vutla Naga Sai Akhil, 41
Y
Yogesh Kakde, 281
