A Survey on Data Stream Mining Towards the Internet of Things Application

Full text

A Survey on Data Stream Mining Towards the
Internet of Things Application
* Saifun Nahar
1
, Ting Zhong
1
, Happy N. Monday
2
, Michael O. Mills
2
, Grace U. Nneji
1
, Hassan S. Abubakar
3
1
School of Information and Software Engineering
2
School of Computer Science and Engineering
3
School of Information and Communication Engineering
University of Electronic Science and Technology of China
Chengdu, Sichuan P.R China
* sainahbd@163.com
Abstract
—
In the era of the Internet of Things, a widespread of
applications depend on time with the various number of different
data generated and collected from different devices available.
These devices depend on the type of application. These fast
stream data are real-time and large in dimension for the purpose
of making decision as well as predicting future occurrence and
analytics. Datastream analytics of internet technology for both
businesses and everyday life is very valuable in terms of
developing good quality of life. In this study, first of all, we focus
on the concept of Internet of Things and its relationship with its
architecture, large and flowing data. In addition, the approach of
Internet of Things applied knowledge discovery process and deep
learning frameworks are presented in this paper. Finally, the
Internet of Things and its features are introduced in this work as
well as the commonly used tools.
Keywords—internet of Things; Big Data Analytics; Deep Learning;
Flowing Data Mining; Data Processing Platforms
.
I. I
NTRODUCTION
Internet of Things can be viewed in a unique way where
objects establish communications with themselves in a
common worldwide network. Objects in this network interact
with each other using a specific communication protocol or
various communications protocols [1] [2]. According to
research, 10-11 billion devices are connected to the internet
presently. It is predicted that the number of gadgets connected
on the internet will increase to 50 billion by 2020. According
to the same research, in 2003 the ratio of interconnected
gadget to a person was 0.08. The 2020 estimate is 6.48. Also
in 2020, the traffic information generated by the typical
household appliance produced in 2008 was 50% bringing to a
huge trillion GB of internet traffic in approximation [3].
Internet of Things exposes information hidden in its large data
so as to improve the quality of lives by removing complex task
[4]. To subdue the difficult and complex task associated with
the traditional methods, new inferences and learning
approaches technologies, algorithms, infrastructures are
required. Fortunately, fast data processing and more developed
machine learning techniques improvements enable large data
analytics and information extraction [5]. Beyond big data
analysis, Internet of Things data with high-speed data streams
and time accuracy actions developed to support applications
with an analytical class, fast and flowing data analytics
concept was formed. Some researchers have found cloud
infrastructures and service analytics frameworks and
approaches. Unnecessary communication closes the data
source to avoid delays. [6] [7]. In this study, the importance of
data analysis in the Internet of Things in terms of the
techniques and tools used are summarized. Deep learning
application is also introduced in the rest of the study.
II. T
HE
I
NTERNET
O
F
T
HINGS
Wireless communication is becoming increasingly
widespread. This the simplest idea of the concept, specific
addressing RFID, sensors, electronic labels, etc. Around us,
different objects interact with each other and also collaborate
for common goals [3]. The Internet of Things structure is
attached to each other via the Internet. A small group of
objects can connect with other objects. Therefore the
architecture must be flexible. The simplest form of studies
done on the Internet of Things is; detection layer, network
layer, and application layer. It forms the new platform for the
Internet of Things over the years. Layer views are specified an
architecture widely used is five layers [3];
• Sensors and actuators in the Object Detection
Layer obtained by different devices via information
such as temperature, humidity, weight, speed,
acceleration, and location brought together.
Configure different types of objects. The plug and
play mechanism is used in this layer.
• From the Detection Layer of the Object,
Abstraction Layer secures the Service. Management
Layer of the resulting data communications via
channels.
• Service Management Layer. Address and name of
the service is the layer where it is paired with
requests. Internet of Things program is written to
support an application with different specific
hardware platforms interacting with objects of
different types.
The 2019 Technology Innovation Management and Engineering Science International Conference (TIMES-iCON2019)
978-1-7281-3755-1/19/$31.00 ©2019 IEEE
Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on January 07,2022 at 06:35:40 UTC from IEEE Xplore. Restrictions apply.

• Application Layer meets the need of the object.
Internet of Things system created for solutions to
meet the layer’s needs. This layer provides
information to request such as the amount of
moisture in the air, temperature.
• The Business Layer is designed to manage the
management layer. Application with the data
obtained from the layer of the business model is the
creation of graphs, tables of this layer responsibility.
A. Big Data and Data Flow in the Internet of Things:
Data volume, data types, variety, and data rate are the
features of big data [8], [9]. For the collected data,
features such as diversity, speed, data volume, reality,
and variability can be counted as value. Internet of
Things flowing data from traditional big data
separating properties are as follows [10]-[12];
• Numerous amounts of data received from devices
are distributed and continue flowing data.
Applications are in real-time and updated
periodically.
• Internet of Things applications are mostly sensors.
For this reason, location and timestamp are available.
• Due to the size of data in the Internet of Things
applications, there are data errors and noise during
transmission and transmission of data.
• Data streams are self-directed amongst objects and
occur in a stochastic process.
• For correct processing of data, transmission order
is quite important.
• System data elements, data flow or data edit
around the streams have corresponding control.
• Once an element is processed from the data stream,
it is either removed or archived. Most especially, the
memory cannot be restored if not stored.
• To query the data flow in a traditional relational
model is quite different
• Unlike stack data, data must be processed.
III. I
NFORMATION
D
ISCOVERED
O
N
T
HE
I
NTERNET
Getting a large amount of data on the internet is called big
data. To keep this data while analyzing, revealing hidden
patterns becomes difficult. Objects Information discovery
process of data obtained in the internet environment with the
implementation of interesting and unknown information using
traditional machine learning or deep learning method with
learning characteristics [4]. Some studies show how well
machine learning algorithms have been applied to data stream
mining. However, it is a topic of intensive study. There are
many proposed approaches and algorithms [13]. Deep learning
algorithms have been proposed by many researchers in data
stream mining due to its learning new features and use it for a
classification task. These techniques are considered problem-
oriented but may not be the same depending on the approaches
and nature of data [6].
Fig 1: Machine Learning VS Deep learning
Fig 2: Deep Learning Architecture
Deep learning is self-learning architecture of neural network
[23]. A deep learning algorithm is capable of learning features
and extraction. Each successive layer takes the output from the
preceding layer as input [14] [15]. Different frameworks are
available for the use of deep learning architectures in different
areas. Each of these frameworks supports deep learning
architecture, and optimization algorithms; depending on the
development, ease of use and advantages [16].
Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on January 07,2022 at 06:35:40 UTC from IEEE Xplore. Restrictions apply.

Table 1: Properties of Deep Learning Frames
Tools Core
Language
Interface
Benefits Drawbacks
H2O Java R, Python, Scala,
REST API
Wide Range Of Interface A limited number of
models are supported.
It is not flexible.
Tensor flow C++ Python, Java, C,
C++, Go
Long Short Term Memory(LSTM) is quite fast.
Support Visualization Network.
More Python-based
frameworks are slowly
trained.
Theano Python Python Support various model. On the GPU(Graphics
Processing Unit) LSTM is quite fast.
Many low-level
Applications
It has Program Interface.
Torch Luna C, C++ Support various model. Good documentation.
The debugger has debugged messages.
It is open to learning a
new language.
Caffe C++ Python, MATLAB Provides a collection of reference models.
Easy to change platform.
Has effective results in folded networks
It is not very effective in
iterative networks.
Neon Python Python It has a fast training time. Easy platform adaptation
and supports modern architectures
CPU (Central Processing
Unit) multiple uses.
Chainer Python Python Supports modern architectures. Complex
architectures can be easily placed. The dynamic
model can be changed.
Slower forward in some
scenarios calculation
feature.
IV. D
ATA
A
NALYTICS
P
LATFORMS
In accordance with the needs of organizations, Internet of
Things plays a huge role in real-time data collection for
analytics platforms. The number of data produced has doubled
in size due to increased usage of devices such as phones,
sensors, etc. Data analytics platform features [18] summarized
in the study;
• Analyze flowing data in real-time and should be
capable of reporting.
• Users regardless of where they are located should
be able to login to the platform.
• Independent usage of interfaces should be designed
as easy.
• Be able to perform various types of data and data
should be easy to adapt.
TABLE 2: Features of Flowing Data Analytic Tools
Flowing Data Analytics
tools
Features
Stream Analytics Integrates many machines into one platform.
Infrastructure technology does not require much user attention
Information Provides off-pack services.
Keeps all data connected and eliminates the hassle of manual code writing
.
SAP Event stream processor Displays future event flows using trends, patterns, and a correlation between them.
Uses notifications and alarms to keep you informed of opportunities and threats.
Oracle Stream Analytics Includes interrelated Visual Geoprocessing and Geoforce for spatial analytics. New Expressive Patterns with
machine learning capabilities for spatial, statistical, anomaly detection library
.
SQL Stream It is a distributed platform developed according to open standards.
Flexible development, scalability, analytical operations and reuse of applications through the application
program interface features.
Apache Flink It has limited and unlimited data processing capacity. Run applications in any environment and scale.
Power transfer feature is available in the memory performance.
It ensures accurate results even in irregular, late data.
High tolerance to error.
It can work efficiently on thousands of nodes and manage latency.
In addition to data-driven windowing, it has flexible windowing capability based on numbers and sessions.
Applications can update or process historical data without any loss and interruption.
The application program interface is very useful.
User-friendly and developed to cover all common jobs.
Configuration does not require the use of memory, network, serializer.
Spark Hadoop can work on Mesos, in the cloud or alone. Different distributed data such as HDFS (Hadoop
Distributed File System), Cassandra, HBase, S3 resources. It empowers different library stacks such as Mlib,
Dataframes, SQL, GraphX and Spark Streaming.
Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on January 07,2022 at 06:35:40 UTC from IEEE Xplore. Restrictions apply.

V. M
ODE OF
W
ORK
Data are generated from meaningful yet different sources
for the application of IoT technology. Internet of Things is
playing a major role in today’s world. Its impact and
technology vary across different applications such as vehicle
tracking, parking systems, payment system and utility
management system with sensors designed to emit rays. These
sensors are cameras that sense motion, sound as well as an
emitting ray. The pressure sensor is sensor cameras that emits
infrared rays to sense pressure [24]. These sensor utilizes
similar technology by sending short messages as SMS and
Necessary warnings to users [19]. With the progress of
wearable technologies, the monitoring of the health
parameters of individuals has facilitated wearable sensors
capable of reading blood pressure, heartbeats, and sleeping
patterns. Based on this technology, the data obtained can be
used to prescribe a drug to patients with having to schedule a
physical meeting. This technology can also prevent patients
from taking the wrong dose of drugs at the wrong time [20].
Visually Impaired Navigation System is designed for people
with disabilities to help them while shopping. The
supermarket is divided into cells radio placed in frequency
tags and the navigating tags are mapped [21]. Individuals
parsing the sound of the smartphone and using the recognition
system to go to the phone notifies and go the person with
Bluetooth and WLAN technologies routing for the hearing
impaired temperature sensor taken with temperature sensor
reached through a status check to a control center. Disable
individual with shining light or vibration warning is provided
[20] [22]. Smart environment Sensors placed in homes and
workplaces to measure ambient temperature, humidity, and
light for personal preference. These parameters can be
adjusted according to weather conditions. Data obtained by
working on peoples' personal habits adjustments are possible
[3]. Smart city applications for garbage separation, traffic
control, energy network as well as forecasting the air quality
[16] [24].
VI. C
ONCLUSION
Internet of Things, deep learning and flowing data mining
concepts have contributed positively to our lives, society, and
the world as a whole. Big data research has become quite
popular amongst researchers nowadays. Internet of Things
technology through the advent of big data has changed the
way we perceive information. Stream data mining offers a
platform for real-time data analytics. The impact of the
application of data stream mining towards IoT can not be
overemphasized due to its numerous advantages. Data
obtained are used for decision making and prediction of
events. Finally, This paper has presented to its readers the
numerous benefits and application of data stream mining
towards IoT.
A
CKNOWLEDGMENT
Work supported by University of Electronic Science and
Technology of China.
R
EFERENCES
[1] C.W. Tsai, C.F. Lai, M.C. Chiang, L. T. Yang, “Data Mining for the
Internet of Things: A Survey.” IEEE Communications Surveys and
Tutorials, vol. 16, pp.77-97, November 2013.
[2] M. A. Alsheikh, S. Lin, D. Niyato, and H.P. Tan, “Machine Learning in
Wireless Sensor Networks: Algorithms, Strategies, and Applications,”
IEEE Communications Surveys & Tutorials, vol. 16, pp.1996-2018,
April 2014.
[3] M. Mazhar Rathore, Anand Paul, Won-Hwa Hong, Hyun Cheol Seo,
Imtiaz Awan, Sharjil Saeed, “Exploiting IoT and Big Data Analytics”,
Sustainable Cities and Societies, vol.40, pp.600-610, July 2018.
[4] Atzori L., Lera A., “Internet of Things: Survey”, Computer Networks,
vol.54, pp.2787-2805, June 2010
[5] M. Zaharia, M. Chowdhury, T. Das, A. Dave, J. Ma, M. Mccauley, M.
Franklin, S. Shenker, and I. Stoica, “Fast and Interactive Analytics Over
Hadoop Data with Spark”, USENIX Login, vol.37, no.4, August 2012.
[6] C. Engle, A. Lupher, R. Xin, M. Zaharia, M. J. Franklin, S. Shenker, and I.
Stoica, “Shark: Fast Data Analysis using Coarse-grained Distributed
Memory”, Proceedings of 2012
ACM SIGMOD International Conference on Management of Data.
ACM, 2012.
[7] M. Chen, S. Mao, Y. Zhang, and V. C. Leung, “Big Data: Related
Technologies, Challenges, and Future Prospects”, Springer, pp.19-32,
July 2015.
[8] W. Fan, A. Bifet, “Mining Big Data: Current Status and Forecast to The
Future”, ACM sIGKDD Explorations Newsletter, vol.14, pp.1-5,
December 2012.
[9] H. Hu, Y. Wen, T.S. Chua, and X. Li, “Toward Scalable Systems for Big
Data Analytics: A Technology Tutorial”, IEEE Access, vol.2, p.652-
687, June 2014.
[10] Y. Demchenko, P. Grosso, C. De Laat, and P. Membrey, “Addressing
Big Data Issues in Scientific Data Infrastructure”, International
Conference on Collaboration Technologies and Systems (CTS), July
2013.
[11] Sorour S., Mohammedi M., “Deep Learning for IOT Big Data and
Streaming Analytics: A Survey”, IEEE Communications Surveys &
Tutorials, vol.20, pp.2923-2960 June 2018.
[12] Guoqiang Z., Li-Na W., Ling X., Dong J., “An Overview on Data
Representation Learning: From Traditional Feature Learning to Recent
Deep Learning”, The Journal of Finance and Data Science, vol.2,
pp.265-278, December 2016.
[13] Z. Fadlullah, F. Tang, B. Mao, N. Kato, O. Akashi, T. Inoue, and K.
Mizutani, “State-of-the-art Deep Learning
[14] S. Han, J. Pool, J. Tran, and W. Dally, “Learning Both Weights and
Connections for Efficient Neural Network” in
Advances in Neural Information Processing Systems, vol.1, pp.1135-
1143, December 2015.
[15] Xiaofeng X., Siping L., “IoT Data Analytics Using Deep Learning”,
IEEE, 2016.
[16] Ching Yu Chen, Jui Hsi Fu, Today Sung, “Complex Event
Processing for the Internet of Things and Its Applications”, IEEE,
vol.20, pp.2923-2960, August 2014.
[17] Top 20 Free Open Source and Premium Stream Analytics Platforms.
(2018, 1 Eylül)
[18] C.W. Tsai, C.F. Lai, M.C. Chiang, and L. T. Yang, "Data Mining for the
Internet of Things: A Survey", IEEE Communications Surveys &
Tutorials, vol.16, pp.77-97, November 2013.
Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on January 07,2022 at 06:35:40 UTC from IEEE Xplore. Restrictions apply.

[19] Y. J. Fan, Y. H. Yin, L. Da Xu, Y. Zeng, and F. Wu, "IOT Based Smart
Rehabilitation System", IEEE Transactions on Industrial Informatics,
vol.10, pp.1568-1577, January 2014.
[20] M. Price, J. Glass, and A. Chandrakasan, “A Scalable Speech
Recognizer with Deep-Neural-Network Acoustic Models and Voice-
Activated Power Gating”, Proceedings of the IEEE ISSCC2017, March
2017.
[21] M. C. Domingo, "An Overview of The Internet of Things for People with
Disabilities,", Journal of Network and Computer Applications, vol.35,
pp.584-596, March 2012.
[22] N. D. Lane, S. Bhattacharya, P. Georgiev, C. Forlivesi, and F. Kawsar,
“An Early Resource Characterization of Deep Learning on Wearables,
Smartphones and Internet of Things Devices”, Proceedings of the 2015
International Workshop on Internet of Things towards Applications,
pp.7-12, November 2015.
[23] I. Kok, M.U. Simsek, S. Ozdemir, “A Deep Learning Model for Air
Quality Prediction in Smart Cities”, IEEE International Conference on
Big Data, December 2017.
[ 24] S.E. Bibri, “The IoT for Smart Sustainable Cities of The Future: An
Analytical Framework for Sensor-based Big Data Applications for
Environmental Sustainability”, Sustainable Cities and Society, vol.38, pp.253,
April 2018.
.
Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on January 07,2022 at 06:35:40 UTC from IEEE Xplore. Restrictions apply.