ChapterPDF Available

FinTech Risk Management and Monitoring

March 2023

March 2023

DOI:10.1007/978-3-031-18552-6_1

In book: Novel Financial Applications of Machine Learning and Deep Learning (pp.3-16)

Authors:

Ariful Hoque

Murdoch University

The recent emergence of financial technology innovations in the financial services industry also faces many challenges due to some significant risks. This chapter aims to identify specific fintech risks and appropriate ways to manage the risks. A qualitative research method is used to explore the objectives of this study. The findings of this study include the major risks behind the rapid development of fintech, and the fintech risk management steps. Also, this study identifies four key regulatory techniques that have important applications in managing and monitoring fintech risks. Finally, the findings summarize the main challenges of fintech risk management.

Framework of data collection and processing

…

Model of an artificial neuron

…

Study model

…

Flow chart of the proposed outlier modification algorithm

…

+43

Proposed methodology for predicting the price of crude oil

…

Figures - uploaded by Ariful Hoque

Content may be subject to copyright.

Content uploaded by Ariful Hoque

Content may be subject to copyright.

International Series in

Operations Research & Management Science

MohammadZoynulAbedin

PetrHajekEditors

Novel Financial

Applications

ofMachine

Learning and Deep

Learning

Algorithms, Product Modeling, and

Applications

International Series in Operations Research &

Management Science

Founding Editor

Frederick S. Hillier, Stanford University, Stanford, CA, USA

Volume 336

Series Editor

Camille C. Price, Department of Computer Science, Stephen F. Austin State Uni-

versity, Nacogdoches, TX, USA

Editorial Board Members

Emanuele Borgonovo, Department of Decision Sciences, Bocconi University,

Milan, Italy

Barry L. Nelson, Department of Industrial Engineering & Management Sciences,

Northwestern University, Evanston, IL, USA

Bruce W. Patty, Veritec Solutions, Mill Valley, CA, USA

Michael Pinedo, Stern School of Business, New York University, New York, NY,

USA

Robert J. Vanderbei, Princeton University, Princeton, NJ, USA

Associate Editor

Joe Zhu, Foisie Business School, Worcester Polytechnic Institute, Worcester, MA,

USA

The book series International Series in Operations Research and Management

Science encompasses the various areas of operations research and management

science. Both theoretical and applied books are included. It describes current

advances anywhere in the world that are at the cutting edge of the ﬁeld. The series

is aimed especially at researchers, advanced graduate students, and sophisticated

practitioners.

The series features three types of books:

•Advanced expository books that extend and unify our understanding of partic-

ular areas.

•Research monographs that make substantial contributions to knowledge.

•Handbooks that deﬁne the new state of the art in particular areas. Each

handbook will be edited by a leading authority in the area who will organize a

team of experts on various aspects of the topic to write individual chapters. A

handbook may emphasize expository surveys or completely new advances (either

research or applications) or a combination of both.

The series emphasizes the following four areas:

Mathematical Programming: Including linear programming, integer program-

ming, nonlinear programming, interior point methods, game theory, network opti-

mization models, combinatorics, equilibrium programming, complementarity

theory, multiobjective optimization, dynamic programming, stochastic program-

ming, complexity theory, etc.

Applied Probability: Including queuing theory, simulation, renewal theory,

Brownian motion and diffusion processes, decision analysis, Markov decision

processes, reliability theory, forecasting, other stochastic processes motivated by

applications, etc.

Production and Operations Management: Including inventory theory, produc-

tion scheduling, capacity planning, facility location, supply chain management,

distribution systems, materials requirements planning, just-in-time systems, ﬂexible

manufacturing systems, design of production lines, logistical planning, strategic

issues, etc.

Applications of Operations Research and Management Science: Including

telecommunications, health care, capital budgeting and ﬁnance, economics, market-

ing, public policy, military operations research, humanitarian relief and disaster

mitigation, service operations, transportation systems, etc.

This book series is indexed in Scopus.

Mohammad Zoynul Abedin •Petr Hajek

Editors

Novel Financial Applications

of Machine Learning

and Deep Learning

Algorithms, Product Modeling,

and Applications

Editors

Mohammad Zoynul Abedin Petr Hajek

Department of Finance, Performance and Faculty of Economics and Administration

Marketing University of Pardubice

Teesside University International Business Pardubice, Czech Republic

School, Teesside University

Middlesbrough, UK

ISSN 0884-8289 ISSN 2214-7934 (electronic)

International Series in Operations Research & Management Science

ISBN 978-3-031-18551-9 ISBN 978-3-031-18552-6 (eBook)

https://doi.org/10.1007/978-3-031-18552-6

©The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland

AG 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 Switzerland AG

The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Preface

Business risk and uncertainty certainly are the toughest challenge in the ﬁnance

domain faced by many researchers and managers. Such uncertainty thereby initiates

an unavoidable risk factor, which is a fundamental element of ﬁnancial theory. To

the best of our knowledge, the ﬁnancial domain has not been a focused subject-

matter for good ML related books. There is also a scarcity of information about how

ﬁnancial enterprises supervise crisis events and achieve turnaround. In order to ﬁx

the multifarious nature of the ﬁnancial problem, this edited book advocates inter-

disciplinary approaches based on machine learning.

Machine learning is involved in the analysis of large and multiple feature

instances. It principally refers to acquiring knowledge and intelligence (by a com-

puter program) from a processed training example for generating predictions. It deals

with computationally intensive techniques, such as cluster analysis, dimensionality

reduction, and support vector analysis. It is principally the area of computer science

and is already frequently applied in social sciences, ﬁnance and banking, marketing

research, operations research, and applied sciences. Moreover, computational

ﬁnance is a domain of applied computer science that is concerned with practical

issues in ﬁnance. It may be characterized as the study of features, instances, and

learning algorithms applied in ﬁnance. It is an interdisciplinary area that integrates

computational tools with numerical ﬁnance. Furthermore, computational ﬁnance

applies arithmetical proofs that can be ﬁtted to economic experiments, thereby

The Novel Financial Applications of Machine Learning and Deep Learning: Algo-

rithms, Product Modelling, and Applications presents the state of the art of the

application of machine learning (ML) and deep learning (DL) in the domain of

ﬁnance. We will present a combination of empirical evidence to diverse ﬁelds of

ﬁnance so that this book is useful to academics, practitioners, and policymakers who

are looking to train novel and the most advanced machine learning classiﬁers. Thus,

the purpose of this book is to provide a broad area of applications to different

ﬁnancial assets and markets. Furthermore, from an extensive literature assessment,

it is evident that there are no existing textbooks that narrate ML and DL to unlike

areas of ﬁnance or to an extensive range of products and markets.

vi Preface

contributing to the advancement of ﬁnancial data modeling techniques and systems.

These computational techniques are utilized in ﬁnancial risk management, corporate

bankruptcy prediction, stock price prediction, and portfolio management. Finally,

this proposed textbook could play an important role in ﬁnancial data learning.

Besides, this volume will be a basis for empirical and theoretical practices. The

empirical experiments aim to minimize ﬁnancial risk and uncertainty by covering

and ﬁtting the most advanced and novel machine learning algorithms. Moreover, it

generates academic literature as well as ﬁnancial product and ﬁnance modeling

inferences toward customer credit risk assessment, data mining, pattern recognition,

bankruptcy prediction, and so on. To be speciﬁc, the volume is broadly divided into

three parts, with the ﬁrst set of chapters focusing on the recent trend and issues of

ﬁnancial technology (FinTech). The second set of chapters comprises empirical

essays on the prediction and forecasting ﬁnancial risk by applying ML and DL

tools and techniques. The third set of chapters combines empirical evidence of

ﬁnancial time-series data forecasting. The volume ends with a set of emerging

technologies in ﬁnancial education and healthcare and their empirical applications.

Part 1: Recent Developments in FinTech

The ﬁrst part presents four chapters on recent development in FinTech.

Chapter “FinTech Risk Management and Monitoring”focuses on risk manage-

ment and monitoring in FinTech. The recent emergence of ﬁnancial technology

innovations in the ﬁnancial services and some signiﬁcant risks are investigated using

the qualitative research method. Additionally, the appropriate way to mitigate the

risk is discussed in this chapter. Besides this objective, this chapter discusses the

major risk behind the rapid development of ﬁntech and the steps for ﬁntech risk

management. The four key regulatory techniques that have important applications in

FinTech management and monitoring are added, and, ﬁnally, the chapter summa-

rizes the main challenges of FinTech risk management.

Chapter “Digital Transformation of Supply Chain with Supportive Culture in

Blockchain Environment”explores the inﬂuence of blockchain on the digital trans-

formation of Supply Chain Management (SCM). This chapter is also aimed to

determine the importance of supportive culture in the adoption of blockchain in

supply chains. The study ﬁndings indicate that the digitalization of supply chain

management by adopting blockchain technology is positively correlated with orga-

nizational prosperity. The chapter also indicates that supportive culture is crucial to

practicing blockchain technology. This study suggests that policymakers and stake-

holders ensure a supportive culture to establish a traceable, efﬁcient, and effective

supply chain.

Chapter “Integration of Artiﬁcial Intelligence Technology in Management

Accounting Information System: An Empirical Study”conducts an empirical

study on the integration of artiﬁcial intelligence technology in management account-

ing information systems. This study established an artiﬁcial neural network-based

Preface vii

model to predict management information and verify the accuracy of the model

using some real data. Five dimensions are considered to develop the model,

accounting analysis management system, accounting decision support system, per-

formance management information system, risk management information system,

and environment management information system.

The essentiality to analyze big data in accounting and ﬁnance is discussed in

Chap. “The Impact of Big Data on Accounting Practices: Empirical Evidence from

Africa”. Evidence indicates that big data signiﬁcantly impact accounting and

auditing accounting, utilizing the diversity of data volume, data variety, and data

velocity. Chapter “The Impact of Big Data on Accounting Practices: Empirical

Evidence from Africa”shows the impact of big data on accounting practices, and

the study area is Africa. The main goal of this chapter is to explore the impacts of big

data on accounting using accountants in Nigeria. Multiple regression is used for

151 responses, and samples are collected using the random sampling method. This

study proves that big data positively and signiﬁcantly affect ﬁnancial reporting,

performance measurement, corporate budgeting, audit evidence, risk management,

and fraud management. This study helps accountants, prospective accountants, and

accounting graduates in their studies.

Part 2: Financial Risk Prediction Using Machine Learning

The second part contains four chapters that discuss the applications of ML and DL

approaches to predict and forecast ﬁnancial risk.

Chapter “Using Outlier Modiﬁcation Rule for Improvement of the Performance

of Classiﬁcation Algorithms in the Case of Financial Data”discusses how to

improve classiﬁer performance by mining and modifying outliers of ﬁnancial

datasets. This chapter offers insights into the Financial Decision Support System

for ﬁnancial decision makers. This study employs four distinct classiﬁcation algo-

rithms such as linear discriminant analysis, k-nearest neighbor, naïve Bayes, and

support vector machine for both original and modiﬁed datasets to detect credit card

fraud. The study’sﬁndings show that the classiﬁers perform better on modiﬁed

datasets than on original credit card datasets.

Chapter “Default Risk Prediction Based on Support Vector Machine and Logit

Support Vector Machine”is a predictive analysis of the machine learning algorithm

for default risk prediction. This study proposes a LogitSVM model that hybridized

the traditional support vector machine with popular logistic regression to assess the

credit default risk. The authors use real-world credit databases to validate the

probability and value of the proposed model. Type I error, type II error, and root

mean square error (RMSE) are used to evaluate the performance of the regressors.

Empirical ﬁndings show that the proposed hybrid model is superior to maximize

accuracy and minimize RMSE. This chapter helps stockholders develop a wide

variety of approaches to predict the credit customers’default risk.

Chapter “Predicting Corporate Failure Using Ensemble Extreme Learning

Machine”shows the corporate failure prediction using the Ensemble Extreme

Learning Machine. The claim is that the early-stage prediction of corporate failure

viii Preface

is essential for banks and ﬁnancial institutions to solve ﬁnancial decision-making

problems. Newly developed artiﬁcial intelligence technique Extreme Learning

Machine has an extremely fast learning classiﬁer. To prove the superiority of this

method, the authors compare the result with four benchmark ensemble methods,

namely multiple classiﬁers, bagging, boosting, and random subspace. Experimental

results on French ﬁrms indicated that bagged and boosted extreme learning machines

showed the best-improved performance.

Chapter “Assessing and Predicting Small Enterprises’Credit Ratings: A

Multicriteria Approach”focuses on small enterprises; it assigns and predicts the

small enterprise’s credit rating using a multicriteria approach. In reality, small

enterprises have made it difﬁcult for ﬁnancial institutions such as commercial

banks to accurately determine the credit risk, creating salient loan difﬁculties due

to short time, high frequency, urgent demand for credit, and a small number of their

loans. To solve this issue, the chapter develops a new approach for assessing credit

risk in small enterprises by combining high-dimensional attribute reduction methods

with fuzzy C-means to grade the credit ratings of enterprises requesting loans.

Part 3: Financial Time-Series Forecasting

The third part contains two chapters that explore empirical evidence of time-series

data modeling.

Chapter “An Ensemble LGBM (Light Gradient Boosting Machine) Approach for

Crude Oil Price Prediction”is on the prediction of crude oil prices. Every second

counts when governments, businesses, and individuals need to know what the future

of the crude oil market will bring in terms of pricing. Estimating the future cost of

crude oil is a crucial step toward building an economy that can last. In order to

effectively predict future crude market prices, this research will use machine learning

and ensemble learning techniques. The model using light gradient boosting (LGBM)

is proposed by the authors to predict the price of crude oil. By analyzing and

modeling the Brent time-series crude oil data, the accuracy and precision of our

predictors can be improved. The LBGM forecast is compared to the lasso regression,

random forest regression, and decision tree regression methods. The results achieved

by the suggested model are quite similar to and better than those obtained by the

baseline model when measured using RMSE, mean absolute percentage error

(MAPE), mean squared error (MSE), and mean absolute error (MAE).

Chapter “Model Development for Predicting the Crude Oil Price: Comparative

Evaluation of Ensemble and Machine Learning Methods”also shows the prediction

of crude oil prices using different methods. This study shows a comparative study of

ensemble algorithms and machine learning algorithms to ﬁnd the best forecasting

model. This research uses machine learning and an ensemble algorithm to forecast

crude oil prices, and it compares the efﬁcacy of three different regression models—

AdaBoost, Bagging Lasso, and Support Vector Regression—to conclude which is

the most suitable. Time-series data on crude oil prices are analyzed and used to

validate the forecasting model. The results of the various algorithms are compared

Preface ix

using an actual vs. anticipated curve. According to the results, the ensemble

AdaBoost method has superior performance. The mean square error, mean absolute

error, root mean square error, mean absolute percentage error, variance score, and R2

are used to verify the outcome. This research will help those with a stake in the crude

oil industry decide and craft policies based on projected future prices.

Part 4: Emerging Technologies in Financial Education

and Healthcare

The fourth part contains three chapters that explore the ﬁnancial education and

healthcare issues and their emerging trends.

Chapter “Discovering the Role of M-Learning Among Finance Students: The

Future of Online Education”investigates the role of m-learning among ﬁnance

students and the future of online higher education. This study aims to ﬁnd the hidden

issues of m-learning in ﬁnance studies. This study is mainly a qualitative approach,

and the ﬁndings show that digitalized education provides the opportunity for major

ﬁnance students to access ﬁnancial markets using the Internet and gain personal and

professional knowledge in a better way rather than traditional learning. It also shows

that m-learning has a signiﬁcant positive relationship with the effectiveness of online

education. This analysis has a signiﬁcant implication for education policymakers and

practitioners.

Chapter “Exploring the Role of Mobile Technologies in Higher Education: The

Impact of Online Teaching on Traditional Learning”demonstrates how technolog-

ical evolvements derive the conduction of higher education, especially mobile

technology. This study also intended to detect the factors that attract pupils who

do not adopt an online education system. A qualitative approach is used to determine

the pros and cons of the technology-based education system in universities. The

authors reveal that the adoption of mobile technologies in academic education

enables students to access valuable resources free of cost and effortlessly, which in

turn helps them to develop strong knowledge and understanding of their study

contents. This study opens up a new arena for research scholars to discover the

importance of online education systems.

Chapter “Knowledge Mining from Health Data: Application of Feature Selection

Approaches”assessed the performance of feature selection techniques in knowledge

mining of health datasets. This study compared seven popular knowledge mining

approaches on six popular Affymetrix and cDNA datasets. Employing a support

vector machine classiﬁer, the study determined the knowledge miners’accuracy and

area under the curve values. The ﬁnding of this chapter informs that the simple lasso

knowledge mining algorithm performs well on Affymetrix datasets while random

forest performs well on cDNA datasets. This chapter contributes to the existing

literature by mentioning the state-of-the-art knowledge mining approaches in health

informatics.

To conclude, this edited volume would provide both practical and managerial

implications of ﬁnancial and managerial decision support systems that capture a

wide range of ﬁnancial data traits. It would guide the execution of risk-adjusted

ﬁnancial product pricing systems, supplemented with a signiﬁcant add up to the

x Preface

ﬁnancial literacy of the investigated study. Furthermore, the book could show a

roadmap to master’s degree students and Ph.D. researchers for ﬁnancial data anal-

ysis. In a wider sense, this speciﬁc volume targets an extensive audience, including

academic and professional ﬁnancial analysts. The contents of this book are expected

to be useful to a wide audience involved in forecasting, modeling, trading, risk

management, economics, credit risk, and portfolio management.

Middlesbrough, UK

Pardubice, Czech Republic

Mohammad Zoynul Abedin

Petr Hajek

Contents

Part I Recent Developments in FinTech

FinTech Risk Management and Monitoring ...................... 3

Morshadul Hasan and Ariful Hoque

Digital Transformation of Supply Chain with Supportive Culture

in Blockchain Environment .................................. 1

Shakila Akter, Mohammad Samiul Haque, Ashrafuzzaman Sohag,

Md. Jahangir Alam Siddikee, and Mohammad Zoynul Abedin

Integration of Artiﬁcial Intelligence Technology in Management

Accounting Information System: An Empirical Study .............. 3

Emon Kalyan Chowdhury

The Impact of Big Data on Accounting Practices: Empirical

Evidence from Africa ....................................... 4

Mandella Osei-Assibey Bonsu, Naheed Roni, and Yongsheng Guo

Part II Financial Risk Prediction Using Machine Learning

Using Outlier Modiﬁcation Rule for Improvement of the Performance

of Classiﬁcation Algorithms in the Case of Financial Data ........... 75

Md. Rabiul Auwul, Md. Ajijul Hakim, Fahmida Tasnim Dhonno,

Nusrat Afrin Shilpa, Ashrafuzzaman Sohag,

and Mohammad Zoynul Abedin

Default Risk Prediction Based on Support Vector Machine

and Logit Support Vector Machine ............................ 9

Fahmida-E-Moula, Nusrat Afrin Shilpa, Preity Shaha, Petr Hajek,

and Mohammad Zoynul Abedin

Predicting Corporate Failure Using Ensemble Extreme Learning

Machine ................................................. 10

David Veganzones

Assessing and Predicting Small Enterprises’Credit Ratings:

A Multicriteria Approach .................................... 125

xii Contents

Baofeng Shi

Part III Financial Time-Series Forecasting

An Ensemble LGBM (Light Gradient Boosting Machine)

Approach for Crude Oil Price Prediction ........................ 15

Sad Wadi Sajid, Mahmudul Hasan, Md. Fazle Rabbi,

and Mohammad Zoynul Abedin

Model Development for Predicting the Crude Oil Price: Comparative

Evaluation of Ensemble and Machine Learning Methods ............ 16

Mahmudul Hasan, Ushna Das, Rony Kumar Datta,

and Mohammad Zoynul Abedin

Part IV Emerging Technologies in Financial Education and Healthcare

Discovering the Role of M-Learning Among Finance Students:

The Future of Online Education ............................... 183

Armana Hakim Nadi, Syed Far Abid Hossain, Al Mahmud Hasan,

Mahbuba Rahman Soﬁn, Saadman Shabab, Md. Ahmedul Islam Sohan,

and Chunyun Yuan

Exploring the Role of Mobile Technologies in Higher Education:

The Impact of Online Teaching on Traditional Learning ............ 197

Syed Far Abid Hossain, Armana Hakim Nadi, Rahma Akhter,

Md. Ahmedul Islam Sohan, Faiza Tanaz Ahsan, Mahbuba Rahman Shoﬁn,

Saadmann Shabab, Tanusree Karmoker, and Krishna Paul

Knowledge Mining from Health Data: Application of Feature

Selection Approaches ....................................... 217

Md. Rabiul Auwul, Md. Ajijul Hakim, Fahmida Tasnim Dhonno,

Nusrat Afrin Shilpa, and Mohammad Zoynul Abedin

Part I

Recent Developments in FinTech

FinTech Risk Management and Monitoring

Morshadul Hasan and Ariful Hoque

Abstract The recent emergence of ﬁnancial technology innovations in the ﬁnancial

services industry also faces many challenges due to some signiﬁcant risks. This

chapter aims to identify speciﬁcﬁntech risks and appropriate ways to manage

the risks. A qualitative research method is used to explore the objectives of this

study. The ﬁndings of this study include the major risks behind the rapid develop-

ment of ﬁntech, and the ﬁntech risk management steps. Also, this study identiﬁes

four key regulatory techniques that have important applications in managing and

monitoring ﬁntech risks. Finally, the ﬁndings summarize the main challenges of

ﬁntech risk management.

Keywords Financial technology · FinTech · Risk management · Risk monitoring

1 Introduction

In recent years, substantial development of ﬁnancial technology (Fintech), such as

artiﬁcial intelligence (AI), big data, machine learning (ML), cloud storage,

blockchain, and other technologies, continues to promote the digital transformation

of ﬁnancial institutions (Deloitte, 2019; Hasan et al., 2020a; Wang et al., 2021). The

application of ﬁnancial products and tools is becoming more abundant, and the

efﬁciency and inclusiveness of ﬁnancial services have signiﬁcantly improved. For

example, the popularity of electronic payments, especially mobile payments,

increases the coverage of basic ﬁnancial services. The promotion and application

of ﬁntech have (i) increased the breadth, depth, and speed of ﬁnancial services,

(ii) brought beneﬁts and convenience to users, (iii) helped ﬁnancial institutions

achieve quality and efﬁciency improvements, and (iv) improved the availability of

ﬁnancial services under the new crown epidemic (Hasan et al., 2020b). Given the

M. Hasan (✉) · A. Hoque

Murdoch Business School, Murdoch University, Perth, Australia

e-mail: mohammad.hasan@murdoch.edu.au;a.hoque@murdoch.edu.au

©The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

M. Z. Abedin, P. Hajek (eds.), Novel Financial Applications of Machine Learning

and Deep Learning, International Series in Operations Research & Management

Science 336, https://doi.org/10.1007/978-3-031-18552-6_1

importance of Fintech, most of the positive effects of the rapid development of

consumer ﬁnance in recent years can be attributed to ﬁntech. Such improvements

include enhancing the breadth and depth of encompassing ﬁnancing and industry’s

overall efﬁciency (Hasan et al., 2022; Long, 2016). Traditional ﬁnancial institutions

have found new directions for ﬁnancial service transformation. Fintech transforma-

tion can also play a role in reforming the future economic structure and improving

efﬁciency. At the same time, ﬁntech development carries signiﬁcant downside risks.

For example, the rapid growth of ﬁntech also creates new problems as it solves the

shortcomings of traditional ﬁnancial services. These downside risks often make

things very challenging for the policymakers to enable new opportunities and

safeguard traditional weaknesses. Also, risks impact ﬁntech companies’strategic

goals. Thereby, managing the risks involved in ﬁntech services is one of the essential

jobs of ﬁntech institutions. Fintech institutions usually measure, manage, and mon-

itor ﬁntech risks in different ways. The details of the ﬁntech risk management and

monitoring process are given in the following section of this chapter.

4 M. Hasan and A. Hoque

2Deﬁnition of FinTech

The word Fintech is a synthesis of ﬁnance (Fin) and technology (Tech) (Hasan et al.,

2020b,2021). Fintech is a technology-oriented ﬁnancial innovation that transforms

or innovates ﬁnancial products and business models using the results of modern

science and technology to promote the quality and efﬁciency of ﬁnancial services

(Aggarwal, 2014; Gai et al., 2018; Gomber et al., 2017). Fintech refers to ﬁnancial

innovations provided by technologies, especially AI, Blockchain, big data analytics,

cloud computing, and other means to redesign traditional ﬁnancial products, pro-

cesses, models, and organizational structures (Goldstein et al., 2019; Hasan et al.,

2020a). Fintech services include digital payment, digital investment, crowdlending,

crowdfunding, and online banking. The rise of ﬁnancial technology on a global scale

has signiﬁcantly improved the service level. Also, the operating efﬁciency of banks

has fundamentally changed the banking industry’s original competitive environment.

In response to the rapidly changing competitive environment, banks have already

started their journey with ﬁnancial technology. Banks can use mobile Internet, bio-

metrics, big data, AI, and other technologies to broaden service channels, reduce

manual services, improve ﬁnancial institutions’full-process risk management and

control capabilities, and reduce compliance and additional operating costs.

3 What Is FinTech Risk?

The rise of new Fintech ﬁrms also means some unknown challenges and risks must

be addressed appropriately. Signiﬁcant innovation poses challenges not only for

ﬁnancial institutions but also for regulators. Fintech risk is a threat that arises

during consumer ﬁnancial transactions and dealing through ﬁntech technologies. In

other words, the danger posed by technological innovations when using ﬁnancial

services. Also, ﬁntech risk can be deﬁned as any potential failures, shortcomings,

and misuse of technology that disrupt consumers’ﬁnancial dealings. Fintech risk

includes many hidden risks and contagious issues that are discussed in the following

sections. In addition, some of the risks and challenges are caused by the improper use

of ﬁnancial technology and some problems with ﬁnancial technology itself.

FinTech Risk Management and Monitoring 5

4 Importance of Maintaining FinTech Securities

The upgrade risk supervision and the emergence of new technologies make a big

difference between the future development of risk management and current well-

known risk management capabilities. It is highly essential for ﬁnancial institutions to

reconsider and leverage emerging technologies to change their existing risk man-

agement methods to improve risk management quality and efﬁciency. Also, ﬁnancial

institutions need to consider ﬁnancial technology risks to make the risk management

approaches more dynamic and capable of responding quickly to new development

trends. However, maintaining ﬁnancial security is a fundamental strategic issue

related to one country’s overall economic and social development. And the accurate

judgment of hidden risks is a prerequisite for ensuring ﬁnancial security. Therefore,

for building ﬁnancial power, it is essential to pay attention to and maintain ﬁnancial

security while promoting ﬁnancial innovation and strengthening the prevention of

ﬁnancial risks. In this aspect, using emerging risk management technologies is also

important to improve the quality of risk management.

5 Risks Behind the Rapid Development of FinTech

This study thinks that technology is neutral; the key difference is how and who uses

it. Due to the speciﬁcs of ﬁnancial technology, ﬁnancial institutions have to address

aseries of new risks while dealing with ﬁnancial services. Some signiﬁcant ﬁntech

risks are discussed in the following sections. Barefoot (2020)classiﬁed ﬁntech risk

into different categories. These are loss of privacy, rising risks of fraud and scams,

compromised data security, harmful manipulation of consumer behavior, uses of

data that are non-transparent to both consumers and regulators, and discriminatory

and unfair uses of data and data analytics. Also, Fintech companies entering ﬁnancial

or regulatory sectors lack sufﬁcient knowledge, stability, and operational efﬁciency.

Deloitte has also identiﬁed some of the most signiﬁcant risks involved in ﬁnancial

services after using technology, such as strategic risk, cyber security risk, informa-

tion technology (IT) vendor risk, IT resiliency and continuity risk, data management

risk, response risk, third-party risk, technology operations risk, risk of ineffective

risk management, and IT program execution risk.

Zhentao, (2021, July 28) added

market risk, operational risk, liquidity risk, legal risk, regulatory risk, and credit

risk. Risk may represent itself in various forms; however, this study points to ﬁntech

risks that need to be considered in ﬁntech operations. These risks are cyber-attack,

data privacy risk, data misuse and quality, technical risk, credit risk, market risk,

liquidity risk, and regulatory risk. The stated risks are discussed in the following

section.

6 M. Hasan and A. Hoque

5.1 Cyberattack

One of the most known risks for ﬁnancial technology services is the threat of cyber-

attacks, network intrusions, email phishing, malware, and other hazards (Alhayani

et al., 2021; Khan et al., 2022b; Miao et al., 2022). Different malware and

ransomware can easily corrupt data, disrupt and shut down computing processes,

and cause signiﬁcant ﬁnancial and reputational damage (Ankita & Rani, 2021;

Sharma et al., 2021).

5.2 Data Privacy Risk

Data privacy is one of the most critical concern for ﬁntech industry. Data privacy

risks primarily focuses on customer data theft, which is one of the most burning

issues nowadays. Due to hacking of customer data, such as personal identity

information, bank accounts, and card information, both ﬁntech users and companies

are continuously losing money. Due to the booming expansion of ﬁntech companies,

data privacy concerns are also booming.

5.3 Data Misuse and Quality

Fintech services deal with millions of data every day. Thus, dealing with big data

creates signiﬁcant risks of data misuse and poor data quality (Clarke, 2016). Also,

due to the absence of proper data regulatory standards, in some cases, ﬁntech

https://www2.deloitte.com/us/en/pages/center-for-board-effectiveness/articles/information-tech

nology-risks-ﬁnancial-services.html

https://www.idx.us/knowledge-center/data-privacy-concerns-in-booming-ﬁntech-industry

companies process poor quality data; thereby, the poor quality data raises important

threat to the effective decision-making process (Barefoot, 2020). The misuse of data

is also considered as a breach of data privacy that ultimately damages a ﬁnancial

institution’s reputation regarding data privacy concerns and undermines institu-

tion’s business interest.

FinTech Risk Management and Monitoring 7

5.4 Technical Risks

Since ﬁnancial technology has not yet achieved effective breakthroughs in security

technology, the technical deﬁciencies of ﬁntech and its dependence on information

system will reduce ﬁntech’s security performance and expand the scope of security

challenges. Also, the application of new technology has not received the necessary

risk assessment. As a result, some organizations blindly pursue the so-called sub-

versive technologies without rigorous testing and risk assessment.

5.5 Operational Risk

The Fintech sector integrates the ﬁnancial industry, technology companies, and

market infrastructure operators. In this aspect, its’operation is complex compared

to other sectors. Thereby, in any case of a high concentration of different industries,

ﬁnancial risks may also arise once a risk arises in any of the sectors.

5.6 Credit Risk

Online credit or loan is one of the most popular ﬁntech services. The online loan

business easily causes credit risks or default of borrowers (Bussmann et al., 2020;

Santoso et al., 2020). Traditional ﬁnancial institutions are exposed to the risks posed

by ﬁnancial technology companies. The cooperation between ﬁnancial institutions

and P2P online loans, third-party payments, and crowdfunding have been continu-

ously strengthened. Any irregular cooperation, violations, and inadequate supervi-

sion can easily lead to cause a rise in credit risk. Also, there is a risk of a lack of

borrowers’information compared to traditional banks (Bussmann et al., 2020).

5.7 Market Risk

Fintech has broken through the temporal and spatial barriers that exist between

traditional ﬁnancial institutions, ﬁnancial institutions and non-ﬁnancial institutions,

and between economic entities. When a risk breaks out, it spreads faster and has a

more signiﬁcant impact on ﬁnancial institutions. For example, commercial banks

face unexpected changes due to continuous market transformation. This continu-

ous market transformation also increases the risk of bankruptcy of the commercial

bank (Yao & Song, 2021b). Also, the return from ﬁnancial technology products is

not stable, and the high-yield model that attracts investors is not sustainable in some

cases. These market risks always impact the ﬁnancial stability and performance of

ﬁntech service providers (Li, 2021; Yao & Song, 2021a).

8 M. Hasan and A. Hoque

5.8 Liquidity Risk

The cooperation of ﬁnancial institutions with P2P lending, alternative ﬁnanc-

ing, Internet wealth management, third-party money transfer services, and Internet

banks can easily cause liquidity risks. The market failures cause systematic liquidity

risk in the ﬁnancial market infrastructure (Avgouleas & Kiayias, 2019). Liquidity

risk may occur in different aspects, such as when in P2P online loans use high-

interest rates, it creates unfair market competition. The unfair market competition

also inﬂuences cash management of traditional banks. The unfair competition also

inﬂuences banks’capital chain, thus also causing liquidity risks. Also, once the

ﬁnancial industry experiences major instability, it will cause large-scale difﬁculties

in cashing out funds, which will initiate liquidity risks and interest rate risks (Lee &

Shin, 2018). As a result, this kind of P2P and other online ﬁnancial products with the

characteristics of popularization and network externalities will lead to unpredictable

losses for society.

5.9 Regulatory Risk

Fintech came into the market within a very short period of time with compli-

cated business processess, and the industry legal system has not been established yet.

Therefore, compliance or regulatory risks are more prominent in ﬁntech services. For

example, there are number of blind spots and loopholes in the existing laws,

regulations, and supervision rules in the ﬁnancial industry. The industry’s lack of

legal treatment and supervision basis leads to some illegal businesses. Institutions

use legal loopholes to carry out criminal and unlawful activities, causing economic

losses to ﬁnancial institutions.

FinTech Risk Management and Monitoring 9

6 FinTech Risk Management, Monitoring,

and Applications

Financial institutions must do a good job of monitoring and managing risk while

providing ﬁnancial transactions, product marketing, business handling, and after-

sales service. Handling or managing risks effectively is an important factor in

successful ﬁntech services.

Considering the importance of effective FinTech risk

management, monitoring, and applications, ﬁntech institutions should focus on the

following issues.

6.1 FinTech Risks Management

Usually, the industry should focus ﬁrst on general risk management practices. Later,

they can focus on speciﬁc actions or processes that will help manage ﬁntech risks.

This study focuses on the necessary steps of ﬁntech risk management in the

following section.

6.1.1 Identify and Categorize Fintech Risks

Risk management teams use different tools, such as AI, ML algorithms, and other

technology, to identify ﬁntech risks. Risk analysts should identify when, where,

why, and how ﬁntech risks can occur. Also, it needs to be recognized by both the

internal and external parties involved in the risks. Besides, risk analysts should

identify the parties who might be affected if any risk occurs.

Identifying risks is

the basic ongoing risk management process.

6.1.2 Risks Measurement

Risk measurement refers to determining the probability of risk occurrence and the

likely impact of such risks on the institution. After identifying major and inﬂuential

risks, all the risks should be categorized and placed on a priority list to sort out which

risks ranked ﬁrst and need urgent solution. The responsible team should have a good

understanding of ﬁnancial data analytics techniques to identify and categorize risks.

Risk measurement is one of the most important stages of analyzing risks with

qualitative and quantitative tools (Alvarez-dionisi, 2020).

Stoneburner, G., Goguen, A., & Feringa, A. (2002). Risk management guide for information

technology systems. Nist special publication,800(30), 800–30.

https://www.business.qld.gov.au/running-business/protecting-business/risk-management/prepar

ing-plan/identify

10 M. Hasan and A. Hoque

6.1.3 Risk Mitigation Plan Focused on Anti-Fraud Methods

and Technological Model

Financial institutions need to develop effective risk mitigation plans and procedures

in the third stage. One of the most vital issues for ﬁnancial institutions is to design

effective anti-fraud methods (Fang et al., 2021) based on product characteristics to

prevent application fraud, transaction fraud, and marketing fraud. Additionally, in

order to track external risk situations such as emerging cybercrime or illicit property

trends, ﬁnancial institutions should be prepared with effective risk mitigation plans

and respond on time when risks arise. Also, ﬁnancial institutions need to specify and

build their own technological model that will work to mitigate different risks.

6.1.4 Analysis and Mitigation

Before mitigating the risks, the risk management team analyzes the risks and their

impacts (Ward, 1999). After analyzing the risks, the team will proceed to the risk

mitigation stage. At this stage, the risk management team determines the probable

solution to prevent or manage the risk and implements the technological models and

other effective ways to mitigate the risks. The team should work with the top

priorities and risks that would have the greatest impact compared to others. In

some cases, the team implements immediate action to prevent the risks from

occurring proactively.

6.1.5 Monitor and Supervision the Performance of Models

It is necessary to continuously monitor the risk of the external participant, including

the risk monitoring of the participant itself and the abnormal behavior of the

participant. Also, ﬁnancial institutions must monitor the performance of models

that were built to mitigate the risks. Fintech products often involve big data and

AI models, and some models or algorithms have a problem during rapid execution.

Therefore, continuous monitoring of the model performance is required, such as

carrying out model veriﬁcation in time to check functional efﬁciency to manage

institutional risk.

6.2 Key Regulatory Technology and Applications

The development of ﬁnancial technology supervision is critical. More attention

should be paid to the development of supervision technology in the regulatory

process. There are a number of supervision technologies that have been widely

used in the supervision of banking, securities, insurance, Internet ﬁnance, and other

ﬁelds. Those regulatory technologies are expected to move towards the full-chain

application of ﬁnancial supervision. The industry calls for attention to the develop-

ment of the following regulatory technologies to strict guard against unknown risks

in the development of ﬁnancial technology.

FinTech Risk Management and Monitoring 11

6.2.1 New Encryption Technology

The new encryption security technology is an emerging security tool that can

effectively protect the privacy and ensure the data security of ﬁnancial institution

information. Kaspersky deﬁned data encryption as “Encryption in cyber security is

the conversion of data from a readable format into an encoded format. Encrypted

data can only be read or processed after it’s been decrypted”.

Even in large data

sets, the new encryption technologies can map data objects to a common data

platforms through access control, assisting the regulatory authorities in overcoming

data security issues, and enabling data to be shared with the regulatory authorities.

6.2.2 Blockchain Technology

The powerful function of this technology is manifested in different aspects. It brings

nearly real-time transaction data through smart monitoring (Masuda et al., 2020;

Yang et al., 2022), which allows regulators to more accurately analyze systemic risks

and improve the efﬁciency of on-site and off-site inspections. Also, the transparent

design of blockchain can provide the supervisory authority with direct, instant and

completely transparent, and trustworthy supervisory information (Khan et al.,

2022a) and effectively enhance the supervisory authority’s ability to deal with

ﬁnancial market emergencies.

6.2.3 Machine Learning Technology

Machine learning (ML) technologies provide different services, such as risk predic-

tion, monitoring, and supervision (Abedin et al., 2021a,b; Jordan & Mitchell, 2015;

Mantere et al., 2012). ML tools can use historical data to effectively identify possible

fraud and can be used in the anti-money laundering ﬁeld. It has a unique ability to

stimulate language and text. Once a transaction deviates from compliance require-

ments is found, the system will automatically issue an early warning to ﬁnancial

institutions and regulatory agencies to monitor their transaction (Awoyemi et al.,

2017; Goy et al., 2019; Sunny et al., 2022).

https://www.kaspersky.com.au/resource-center/deﬁnitions/encryption

12 M. Hasan and A. Hoque

6.2.4 Big Data Technology

Big data technology can reorganize and analyze various types of data, obtain

valuable information, and reveal the essential attributes of things. With the aid of

effective analysis and discovery tools, big data allows regulators to brieﬂy see what

has been and is happening in the ﬁnancial market. It can also accurately determine

the probability of upcoming risks, which enhances the supervisor’s ability to allocate

supervisory resources dynamically (Khan et al., 2022c).

6.3 Main Applications of Regulatory Technology

Blockchain, machine learning, big data, and other risk regulatory tools help the

ﬁnancial institution in different aspects, such as smart supervision, fraud detection

and prevention, data management, transaction monitoring, and so on. The major

applications of key regulatory technologies are discussed in the following sections.

6.3.1 Smart Supervision

Regulatory technology uses ML and cloud computing technology to enable the

system to consciously track supervision, identify compliance requirements, provide

targeted response solutions, manage compliance workﬂows, build data reporting

platforms, open up different supervision reports, and other supervision activities.

The Internet generates massive amounts of user data that are difﬁcult to model

manually every day. ML can solve the problem of slow manual model iteration.

For the supervision of ﬁnancial risks, the ML model can efﬁciently and quickly

self-iterate by monitoring the characteristics and performance of the model, loan

groups, and business feedback.

6.3.2 Fraud Prediction and Prevention

Big data helps to ﬁnd clues to illegal activities based on data analysis. For online

transactions, both senders and receivers of the transaction cannot visit physically.

Therefore, this online connection opens room for the applicant for material fraud. In

this case, big data technology can compare the information provided by the applicant

with the authentic and accurate information that has been stored, discover the

difference between the before and after dispatch information and provide evidence

to prevent fraud and crack down on illegal and criminal activities in time. For

example, big data tracks people’s daily trajectories and accurately locates them

based on geographic location. When the applicant’s home address does not match

the registered address or the information, such as the transaction address, is different

from the stored information, the big data system automatically compares and issues

an early warning.

FinTech Risk Management and Monitoring 13

6.3.3 Data Management

The establishment and use of big data technology, cloud computing, and other

platforms are inseparable from data. Raw data is increasingly vital for the accuracy

of risk prediction results. Data management covers using raw data to forecast all

kinds of risk modeling, situation analysis and stress testing, scientiﬁc research and

judgment on various ﬁnancial risks, and formulating solutions. A high-quality

database is needed to accomplish the above things. Therefore, data accuracy,

completeness, and credibility signiﬁcantly impact risk management and improve

risk management performance. With the improvement of data quality requirements,

the operating costs of risk databases also increase accordingly, which puts forward

new requirements for the ability to select data.

6.3.4 Transaction Monitoring

Transaction monitoring is designed to detect unusual behavior that may indicate

the occurrence of other ﬁnancial crimes, such as terrorist ﬁnancing and money

laundering.

Real-time payment transaction monitoring has systemic problems,

such as inaccurate data monitoring, which provides space for money laundering

and other illegal activities. In this aspect, supervisory technology has the character-

istics of intelligent, efﬁcient, and automatic solution generation, which provides the

possibility to discover system defects and eliminate illegal activities. Financial

regulatory authorities use different applications in ﬁnance to improve regulatory

efﬁciency and combat against ﬁnancial crime. Those monitoring and managing

applications prohibit ﬁnancial market’s false transactions and irregularities, and

enhance risk management efﬁciency. Also, regulatory technologies guarantee the

compliance and transparency of transactions and can improve transaction efﬁciency.

7 Challenges of FinTech Risk Management

Today’s business environment is changing rapidly, and risks are also rapidly evolv-

ing. The ﬁnancial industry also faces evolving challenges, such as continuous

regulatory changes, growing awareness of third-party risk, lack of technology

expertise, evolving data governance standards, increasing operational resilience

demands, increasing cybersecurity threats, and other security and data privacy

https://sanctionscanner.com/blog/biggest-transaction-monitoring-challenges-626

issues.

As a result, ﬁntech ﬁrms face complicated risks and compliance challenges.

For example, integrating big data and AI technologies is challenging to implement. It

requires exceptional and high engineering skills and constant costly maintenance.

14 M. Hasan and A. Hoque

In some cases, technological integrations are changing and reshaping the opera-

tions of the ﬁnancial industry.

It is evident that attempting to address these risks

through manual techniques only increases risks, such as the inability to adapt to

regulatory changes, poor data governance, and greater cyber risk. Instead, ﬁntech

organizations may consider taking a more strategic approach to successfully tackle

these difﬁculties.

8 Conclusion

Risk in the ﬁntech industry is a highly concerning issue at present time. Robust and

very effective risk management techniques and strategies are highly demanding. A

sound risk management system makes an organization more dynamic and responds

quickly to emerging threats. This study is one of the ﬁrst to explore hidden risks and

appropriate risk management approaches in the FinTech industry. In addition, this

paper discusses risk monitoring and oversight techniques and their applications to

support the risk management processes. Overall, this research will have a signiﬁcant

implications on the risk management operations of ﬁntech ﬁrms and make a sub-

stantial contribution to the ﬁntech literature.

References

Abedin, M. Z., Hassan, M. K., Hajek, P., & Uddin, M. M. (2021a). Machine learning in ﬁnance and

accounting. In The essentials of machine learning in ﬁnance and accounting (1st edn). Springer.

Abedin, M. Z., Moon, M. H., Hassan, M. K., & Hajek, P. (2021b). Deep learning-based exchange

rate prediction during the COVID-19 pandemic. Annals of Operations Research, 2021,

1. https://doi.org/10.1007/s10479-021-04420-6

Aggarwal, D. V. K. (2014). Financial inclusion in India –Opinion. International Journal of

Commerce, Business and Management (IJCBM), 3(6), 841–849.

Alhayani, B., Abbas, S. T., Khutar, D. Z., & Mohammed, H. J. (2021). Best ways computation

intelligent of face cyber attacks. Materials Today: Proceedings, xxxx.https://doi.org/10.1016/j.

matpr.2021.02.557

Alvarez-dionisi, L. E. (2020). A ﬁntech risk assessment model. ISSCA Journal, 3,1–6.

Ankita, A., & Rani, S. (2021). Machine learning and deep learning for malware and ransomware

attacks in 6G network. In Proceedings - 2021 4th international conference on computational

intelligence and communication technologies, CCICT 2021, pp. 39–44. https://doi.org/10.1109/

CCICT53244.2021.00019

https://www.protechtgroup.com/blog/top-5-risk-management-challenges-for-ﬁntechs

https://www.mobindustry.net/blog/7-key-challenges-ﬁntech-startup-faces-and-their-solutions/

FinTech Risk Management and Monitoring 15

Avgouleas, E., & Kiayias, A. (2019). The promise of Blockchain Technology for global securities

and derivatives markets: The new ﬁnancial ecosystem and the ‘holy grail’of systemic risk

containment. European Business Organization Law Review, 20(1), 81–110. https://doi.org/10.

1007/s40804-019-00133-3

Awoyemi, J. O., Adetunmbi, A. O., & Oluwadare, S. A. (2017, January). Credit card fraud

detection using machine learning techniques: A comparative analysis. In Proceedings of the

IEEE International Conference on Computing, Networking and Informatics, ICCNI 2017,

pp. 1–9. https://doi.org/10.1109/ICCNI.2017.8123782

Barefoot, J. A. (2020). Digital technology risks for ﬁnance: Dangers embedded in Fintech and

Regtech. In M-RCBG Associate Working Paper Series (p. 151). https://www.hks.harvard.edu/

centers/mrcbg/publications/awp/awp151

Bussmann, N., Giudici, P., Marinelli, D., & Papenbrock, J. (2020). Explainable AI in ﬁntech risk

management. Frontiers in Artiﬁcial Intelligence, 3(April). https://doi.org/10.3389/frai.2020.

00026

Clarke, R. (2016). Big data, big risks. Information Systems Journal, 26(1), 77–90. https://doi.org/

10.1111/isj.12088

Deloitte. (2019). Fintech risk and compliance management - A framework to empower the

organization.

Fang, W., Li, X., Zhou, P., Yan, J., Jiang, D., & Zhou, T. (2021). Deep learning anti-fraud model for

internet loan: Where we are going. IEEE Access, 9, 9777–9784. https://doi.org/10.1109/

ACCESS.2021.3051079

Gai, K., Qiu, M., & Sun, X. (2018). A survey on FinTech. Journal of Network and Computer

Applications, 103, 262–273. https://doi.org/10.1016/j.jnca.2017.10.011

Goldstein, I., Jiang, W., & Karolyi, G. A. (2019). To FinTech and beyond. Review of Financial

Studies, 32(5), 1647–1661. https://doi.org/10.1093/rfs/hhz025

Gomber, P., Koch, J.-A., & Siering, M. (2017). Digital ﬁnance and FinTech: Current research and

future research directions. Journal of Business Economics, 87(5), 537–580. https://doi.org/10.

1007/s11573-017-0852-x

Goy, G., Gezer, C., & Gungor, V. C. (2019, March). Credit card fraud detection with machine

learning methods. In UBMK 2019 - Proceedings, 4th International Conference on Computer

Science and Engineering, pp. 350–354. https://doi.org/10.1109/UBMK.2019.8906995

Hasan, M. M., Popp, J., & Oláh, J. (2020a). Current landscape and inﬂuence of big data on ﬁnance.

Journal of Big Data, 7(1), 21. https://doi.org/10.1186/s40537-020-00291-z

Hasan, M. M., Yajuan, L., & Khan, S. (2020b). Promoting China’s inclusive ﬁnance through digital

ﬁnancial services. Global Business Review, 1–23.https://doi.org/10.1177/097215091989534

Hasan, M., Le, T., & Hoque, A. (2021). How does ﬁnancial literacy impact on inclusive ﬁnance?

Financial Innovation, 7(1), 1. https://doi.org/10.1186/s40854-021-00259-9

Hasan, M., Noor, T., Gao, J., Usman, M., & Abedin, M. Z. (2022). Rural consumers’ﬁnancial

literacy and access to FinTech services. Journal of the Knowledge Economy. https://doi.org/10.

1007/s13132-022-00936-9

Jordan, M. I., & Mitchell, T. M. (2015). Machine learning: Trends, perspectives, and prospects.

Science, 349(6245), 255.

Khan, K. M., Arshad, J., Iqbal, W., Abdullah, S., & Zaib, H. (2022a). Blockchain-enabled real-time

SLA monitoring for cloud-hosted services. Cluster Computing, 25(1), 537–559. https://doi.org/

10.1007/s10586-021-03416-y

Khan, M. N. R., Ara, J., Yesmin, S., & Abedin, M. Z. (2022b). Machine learning approaches in

cybersecurity. Data Intelligence and Cognitive Informatics. https://doi.org/10.1007/978-981-

16-6460-1_26

Khan, M. N. R., Tasnim, F., Yesmin, S., & Abedin, M. Z. (2022c). Review of city pricing system

analysis based on big data. Data Intelligence and Cognitive Informatics.https://doi.org/10.

1007/978-981-16-6460-1_25

Lee, I., & Shin, Y. J. (2018). Fintech: Ecosystem, business models, investment decisions, and

challenges. Business Horizons, 61(1), 35–46. https://doi.org/10.1016/j.bushor.2017.09.003

16 M. Hasan and A. Hoque

Li, C. (2021). Quantitative measurement and analysis of FinTech risk in China. Economic

Research-Ekonomska Istrazivanja, 35, 2596. https://doi.org/10.1080/1331677X.2021.1970606

Long, C. (2016). From Fintech to Finlife: The case of Fintech development in China. China

Economic Journal, 9(3), 225–239. https://doi.org/10.1080/17538963.2016.1215057

Mantere, M., Uusitalo, I., Sailio, M., & Noponen, S. (2012). Challenges of machine learning based

monitoring for industrial control system networks. In Proceedings - 26th IEEE International

Conference on Advanced Information Networking and Applications Workshops, WAINA 2012,

pp. 968–972. https://doi.org/10.1109/WAINA.2012.135

Masuda, D., Shinkuma, R., Inagaki, Y., & Oki, E. (2020). Blockchain framework for real-time

streaming data generated in image sensor networks for smart monitoring. In 2020 2nd Confer-

ence on Blockchain Research and Applications for Innovative Networks and Services, BRAINS

2020, pp. 217–221. https://doi.org/10.1109/BRAINS49436.2020.9223311

Miao, Y., Chen, C., Pan, L., Han, Q. L., Zhang, J., & Xiang, Y. (2022). Machine learning based

cyber attacks targeting on controlled information: A survey. ACM Computing Surveys, 54(7),

1. https://doi.org/10.1145/3465171

Santoso, W., Trinugroho, I., & Risfandy, T. (2020). What determine loan rate and default status in

ﬁnancial technology online direct lending? Evidence from Indonesia. Emerging Markets

Finance and Trade, 56(2), 351–369. https://doi.org/10.1080/1540496X.2019.1605595

Sharma, S., Kumar, R., & Rama Krishna, C. (2021). A survey on analysis and detection of Android

ransomware. Concurrency and Computation: Practice and Experience, 33(16), 1–24. https://

doi.org/10.1002/cpe.6272

Sunny, F. A., Khan, M. I., Satu, M. S., & Abedin, M. Z. (2022). Investigating external audit records

to detect fraudulent ﬁrms employing various machine learning methods. Proceedings of the

Seventh International Conference on Mathematics and Computing.https://doi.org/10.1007/

978-981-16-6890-6_38

Wang, R., Liu, J., & Luo, H. (2021). Fintech development and bank risk taking in China. European

Journal of Finance, 27(4–5), 397–418. https://doi.org/10.1080/1351847X.2020.1805782

Ward, S. (1999). Assessing and managing important risks. International Journal of Project

Management, 17(6), 331–336. https://doi.org/10.1016/S0263-7863(98)00051-9

Yang, F., Qiao, Y., Abedin, M. Z., & Huang, C. (2022). Privacy-preserved credit data sharing

integrating Blockchain and federated learning for industrial 4.0. IEEE Transactions on Indus-

trial Informatics.https://doi.org/10.1109/TII.2022.3151917

Yao, T., & Song, L. (2021a). Examining the differences in the impact of Fintech on the economic

capital of commercial banks’market risk: Evidence from a panel system GMM analysis.

Applied Economics, 53(23), 2647–2660. https://doi.org/10.1080/00036846.2020.1864275

Yao, T., & Song, L. (2021b). Fintech and the economic capital of Chinese commercial bank’s risk:

Based on theory and evidence. International Journal of Finance and Economics. https://doi.org/

10.1002/ijfe.2528

Zhentao, Y. (2021, July 28). Risks and countermeasures brought about by the rapid development

of ﬁnancial technology. Qindianstar

Digital Transformation of Supply Chain

with Supportive Culture in Blockchain

Environment

Shakila Akter, Mohammad Samiul Haque, Ashrafuzzaman Sohag,

Md. Jahangir Alam Siddikee, and Mohammad Zoynul Abedin

Abstract This study aims to evaluate the effect and usefulness of digital transfor-

mation of supply chain management (SCM) on blockchain with a supportive culture.

This paper explores the effect of blockchain on SCM under consideration of

automated controls with smart contracts, fundamental attributes, cooperation, sup-

portive culture, transparency and identiﬁcation, and trust building. Here, this study

ﬁnds that the supportive culture has great potential to boost the transformation of

SCM rapidly and successfully. Blockchain technology has the potential to transmit

the supply chain. Finally, this current study indicates that the transformation of SCM

in blockchain with supportive culture has a positive impact on the success of

organizations. Therefore, this study inspires policymakers and stakeholders to

ensure a supportive environment to build a robust sustainable supply chain that

will be traceable, more effective, and efﬁcient.

Keywords Digital transformation · Supportive culture · Supply chain management ·

Blockchain

1 Introduction

Technological or digital transformation is one of the trends that shape the business

world and changes in the work environment. To cope with technological trans-

formations and utilize opportunities that arise from digital technologies, the SCM of

the company faces numerous pressures, such as lack of supportive culture, industry-

speciﬁc guidelines, digital skills, etc. (Agrawal et al., 2020). Digital transformation

S. Akter · M. S. Haque · A. Sohag · M. J. A. Siddikee

Hajee Mohammad Danesh Science and Technology University, Dinajpur, Bangladesh

M. Z. Abedin (✉)

Teesside University International Business School, Teesside University, Middlesbrough, Tees

Valley, UK

e-mail: m.abedin@tees.ac.uk

©The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

M. Z. Abedin, P. Hajek (eds.), Novel Financial Applications of Machine Learning

and Deep Learning, International Series in Operations Research & Management

Science 336, https://doi.org/10.1007/978-3-031-18552-6_2

(DT) is known as a way of developing a new business model that helps organizations

generate relatively greater value (Verhoef et al., 2019). This transformation has an

impact on ﬁrm schedules, capabilities, and business procedures (Da Xu et al., 2018).

DT supports ﬁrms in offering better products and services by eliminating the

obstacles between ﬁnal users, businesses, and objects. A supportive culture is

essential for every business to enhance and share knowledge, learning, resources,

and skills (Bollinger et al., 2002).Organizational culture and environmental sus-

tainability play the driving role in adopting the digital transformation of businesses

by bringing out a continuous change in their structure (Isensee et al., 2020; Khan

et al., 2022). Supportive culture ensures the situation in which human resources can

build a supportive correlation between them, organizational culture, environment,

and their working conditions (Karine, 2020).

18 S. Akter et al.

At present, blockchain technology is practiced in several industries including the

ﬁnance and accounting industry such as capital markets, international trade, corpo-

rate governance, banking, and taxation (Farhana et al., 2022). Blockchain technol-

ogy accelerates consumer conﬁdence by operating transactions more efﬁciently,

traceably, safely, and transparently (Aste & Matteo, 2017; Kshetri, 2018; Queiroz

& Fosso, 2019). Blockchain technology (BCT) represents an appropriately circu-

lated public ledger that covers details about each type of data transaction among

network participants (Singh & Kim, 2018; Yang et al., 2022). Traditional supply

chain management (SCM) has to face a number of problems such as product

tampering, fraud, and, delay, etc. (Petr & Abedin, 2020; Abedin et al., 2020). BCT

has the potential to eliminate the aforementioned difﬁculties through its signiﬁcant

available features, such as anonymity, decentralization, stability, traceability, and

transparency (Ali et al., 2020). The adoption of blockchain in SCM helps increase

the cooperation between supply chain members, efﬁciency in the supply chain

process, and reduce overall cost. To detect and prevent products fraud, blockchain

traceability activities have a signiﬁcant inﬂuence on SCM (Chen, 2018; Sana et al.,

2022). Blockchain has the ability to solve composite issues such as accountability

and transparency (Kshetri, 2018). Therefore, in the perception of SCM, blockchain is

considered as an identical technology (Choi et al., 2020).

Nowadays, digital technology has completely updated how people interact with

their surroundings. Individuals use smartphones, smart watches, personal com-

puters, advanced television units, wearable devices, drones, and self-driving cars

to access and transfer data that are the reﬂection of digitalization (Prasitlumkum

et al., 2020). These technological innovations have a major impact on each sector,

including the supply chain sectors (Abedin et al., 2021). A supply chain is a uniﬁed

system of organizations, people, and information that involves planning, organizing,

controlling, and coordinating the transfer of products and services from the provider

to the consumer (Azzi et al., 2019; Shajalal et al., 2021). Digital technology affects

every phase of human life as well as the supply chain process (Nasiri et al., 2020).

Companies are increasingly aware of these potential developments and strengthen

how the digital supply chain (DSC) can add value to them. DSC is a series of

interrelated actions that are driven by new technology and involved in the supply

chain process (Büyüközkan & Göçer, 2018). DSC can create new forms of revenue

and business value for companies by using various innovative technologies such as

drones, cloud computing, bar code readers, QR codes, and unmanned aerial vehicles

(Bicocchi et al., 2019).

Digital Transformation of Supply Chain with Supportive Culture... 19

Adoption of DT faces plenty of difﬁculties, namely lack of vision, insufﬁcient

leadership knowledge and skills, ﬁnancial inadequacy, and lack of a supportive

organizational culture (Papagiannidis et al., 2020). Therefore, this empirical study

investigates the role of a supportive culture in the adoption of digital transformation,

especially blockchain technology in conducting supply chain activities. This study

tends to detect the potential impact of the blockchain environment and digital

transformation with supportive culture on SCM issues: traceability, transparency,

security, and efﬁciency play. This study contributes to the existing literature on

digital supply chain management and organizational supportive culture. This study

extends the existing domains by identifying the effect of supportive culture in SCM

considering digitalization. This paper suggests that stakeholders consider the orga-

nizational internal and external environment while adopting new technology to carry

out SCM activities.

2 Literature Review

By employing a theoretical framework on archival data from case studies, Kshetri

(2018) explores that blockchain impacts on SCM objectives like quality, reliability,

cost, sustainability, risk minimization, and ﬂexibility. Wang et al. (2019) seek to

identify how BCT changes the traditional supply chain practices. For this purpose,

their study employs narrative analysis and cognitive mapping. Applying transaction

cost theory, the study of Schmidt and Wagner (2019) establishes a preliminary idea

of how blockchain affects supply chain relations. In this regard, they consider

authority decisions and operation costs. Saurabh and Dey (2020) utilize the conjoint

analysis (CA), by developing the theoretical framework, to identify the inﬂuential

factors that affect the BCT in the grape wine supply chain. To detect the ﬁnancial and

operational advantages of adapting blockchain technology rather than a traditional

platform, Giovanni (2020) applies a simple supply chain (SC) model. By combining

the Fuzzy Delphy and Best-Worst method (BWM), Ghasemian et al. (2020) generate

an integrated method to determine the barriers to blockchain adoption in a human-

itarian supply chain management.

The study by Dowty and Wallace (2010) detected the role of organizational

culture in disrupting and restoring the supply chain. In the study by Li et al.

(2016), they explore the organizational pressure to take on Internet-enabled SCM

from the perspective of organizational culture. Conducting survey data from 131 Chi-

nese service and manufacturing ﬁrms, their study develops a conceptual framework

and hypothesis test. By using the mediating effect of structural equation modeling

(SEM), Liou et al. (2012)analyze the institutional commitment in relation to

organizational supportive culture and employee job satisfaction. They collect pri-

mary data from 210 samples of Taiwanese universities. Lin (2013) identiﬁes the

factors for adopting an electronic supply chain management system (e-SCM) from

an organizational, environmental, and technological perspective using logistic

regression. Their survey collects data from 283 managers from Taiwanese ﬁrms.

Conducting questionnaire-based data from 418 graduates from Dutch Business

School (DBS), the Netherlands, Sok et al. (2014) explore the relationship between

work-to-home and organizational culture spillover. Their study utilizes structural

equation modeling and conﬁrmatory factor analysis (CFA).

20 S. Akter et al.

Jabbar et al. (2020) describe the digital transformation of sustainable supply chain

management (SSCM) as big data analytics. They applied a systematic literature

review (SLR) method. Their study considers the Scopus database as article searches

by title, abstract, and keyword. Nasiri et al. (2020) inspect the mediating effect of

smart technologies. In their study, they consider 280 Finnish small and medium-

sized enterprises to show how the organization’s digital transformation affects the

relationship performance from the supply chain perspective. Song et al. (2021) seek

to clarify the various e-commerce methods of the wholesale market that can update

and transform its ecosystem by implementing Information and Communication

Technology (ICT). In their study, they collect 24 interviews as primary data, market

records, papers, internal reports, as well as different published documents as sec-

ondary data from a theoretical point of view. Büyüközkan and Göçer (2018), take

into account the Analytic Hierarchy Process (AHP), Additive Ratio Assessment

(ARAS), and Interval Valued Intuitionistic Fuzzy (IVIF) sets under the Group

Decision Making (GDM) method. Their study initiates a DSC procedure for the

selective activities of suppliers.

On the basis of the literature mentioned above, this study determines the follow-

ing research gap. There are a range of studies dealing with the relationship between

blockchain and supply chain, supportive culture and DT, and DT relations with

SCM, respectively. That means the existing literature covers the interconnections

between corresponding issues, but they cannot reﬂect the impact of a supportive

culture in adopting digital transformation in SCM.

To cover the aforementioned research gap, this study aims to consider the factors

simultaneously. This paper sheds a new light on the importance of a supportive

culture in the face of digital transformation (BCT) in managing supply chain.

3 Methodology of the Study

The method is a description of the manner in which data are collected, analyzed, and

interpreted. This study proposes a systematic literature review (SLR) of academic

and practitioner literature on the Digital Transformation of Supply Chain and

Blockchain technology. This current study conducts several steps of analysis to

include a set of articles in the review. First, for the purpose of this paper, we searched

the top academic journal databases. Accepted articles include the following key-

words blockchain, supply chain transformation, and organizational culture. Second,

this study considers the science citation index (SCI), the social science citation index

(SSCI), and social science citation expanded (SSCE) for papers related to

blockchain, supply chain transformation, and organizational supportive culture.

The time frame of the paper is the data during the 1991–2020 years. But the

maximum data is targeted for the past seven years (2013–2020).

Digital Transformation of Supply Chain with Supportive Culture... 21

Fig. 1 Framework of data collection and processing

Finally, we examine 87 articles including journal article, article in a periodical,

conference proceedings, book chapters, and reports (Fig. 1).

4 Analysis and Interpretation

4.1 Digital Transformation of Supply Chain

Supply chain experts think about how to go forward due to the rapid prosperity of

modern technology. The supply chain process changes a lot with the touch of digital

transformation. To accept these major changes, companies must identify the pros-

pects and barriers made by digital technology. DSC allows ﬁrms to recognize the

customer’s needs, the supplier’s challenges, and continue their operation efﬁciently.

Machine learning (ML)/Artiﬁcial intelligence (AI), blockchain, and Robotic process

automation (RPA) are considered technological assistance to make the ﬁrm digita-

lize (Hartley & Sawaya, 2019); see Fig. 2.

Robotic Process Automation

Robotic Process Automation (RPA) can be deﬁned as a developing technology that

ensures the utilization of software bots to enable the ﬁrm to automate tasks and rule-

based business procedures (Kokina & Blanchette, 2019). Evidence has shown that

about 60% of supply chain experts apply RPA to automate supply chain processes

(APQC, 2018). In DT practice, RPA considers the organizational ﬁrst phase. Data

designs, entry, evaluation, and mining from the Enterprise Resource Planning (ERP)

structure are the main activities associated with RPA (Huang & Vasarhelyi, 2019).

The supply chain conducts various monotonous tasks automatically, such as collec-

tions, operations, and logistics through RPA (Hartley & Sawaya, 2019). Organiza-

tions move forward with digital transformation with RPA for different motives. First,

setting up RPA application is comparatively easy by applying the software bots.

Second, it can be applied faster than a completely reformed process from end to end.

In conclusion, on the basis of business requirements, there are options to improve or

eliminate the capacity.

22 S. Akter et al.

Fig. 2 Supply chain

transformation

Artiﬁcial Intelligence/Machine Learning

Artiﬁcial Intelligence (AI) refers to the potentiality to contribute to engineering and

scientiﬁc assignments by replicating, broadening, and converting human expressions

in an effective and accurate manner (Muthukrishnan et al., 2020). Nowadays, there

are many AI applications in the supply chain and the possibilities of AI applications

are endless. This study considers machine learning (ML) as a subsection of AI for

supply chain operation procedures. ML contains algorithms that can learn compound

operations and develop analytical models from test data (Carbonneau et al., 2008).

Supply chain applications of ML include scheduling of warehouse pick processes,

demand planning, and forecasting, governing the equipment nurturing plans, exam-

ining information to advance the transportation supervision, etc. (Toorajipour et al.,

2021).

Blockchain

Blockchain is an independent digitally Distributed Ledger Technology (DLT)

(Di et al., 2020), holding random information, which is not supervised by a sole or

a corporation of entities; anybody can access this platform easily (Lafourcade &

Lombard-platet, 2020). Although BCT applications were ﬁrst introduced in the

cryptocurrency perspective (Nakamoto, 2008), currently, this technology uses var-

ious contexts such as SCM (Karamchandani et al., 2019), health care record man-

agement (Shi et al., 2020), electronic voting (Nam et al., 2021), the insurance

industry (Kar & Navin, 2021), and so on. Generally, BCT platforms are more secure.

Permitted users have access to include or view particular data. Blockchain adds

positive value in the SCM area in a different way such as product traceability, SCM

transmission, inventory supervision, and customer afﬁliation (Jabbour et al., 2020).

Daily operations are automated using smart contracts through blockchain (Xuan

et al., 2020).

Digital Transformation of Supply Chain with Supportive Culture... 23

4.2 Digital Transformation of Supply Chain in Supportive

Culture

Today’s world is changing a lot by technological innovation. That is why the online-

based or automated business has taken place rapidly rather than a traditional

business. Organizations implement different modern technologies for different rea-

sons, such as meeting customer demand, competitive pressure, and the wide accep-

tance of technology. The digital transformation of supply chains changes the

organizational operation procedure, model, plans, and culture. Digitalization updates

current cultures or creates new ones and uses structures, symbols, and digital art

around the business (Bounfour, 2016).

Based on previous studies, several factors have an impact on supply chain

transformation. Employee engagement and acceptance are considered the most

crucial factors to support the transformation progression (Michela & Burke, 2000).

The supportive culture ensures a collaborative and human-aligned, friendly, moti-

vating, and trustful workplace (Dowty & Wallace, 2010), and it minimizes the

likelihood of negative working experiences for employees by increasing job satis-

faction (Liou et al., 2012). A leader is one who supports and understands the feelings

of others. To successfully implement digital technology, leaders play a major role

(Banks et al., 2019).

Organizational culture refers to the ways in which norms, beliefs, values, and

communications help establish an organization’s emotional and exceptional social

environment (Wu, 2008). Although cultural change is too challenging, any kind of

organizational change culture is crucial. To implement digital transformation, it is

necessary to change strategy, leadership, and organizational culture (Halpern et al.,

2021). Sometimes the chief executive manager and other higher authorities allow the

change. Therefore, the transformation depends greatly on the entire staff support of

the organization. Combining culture and technology is not an easy job, as both

concepts interact with the organizations’subsystems. To adjust the culture in the

digital transformation of the supply chain, a supportive approach is needed. Cabrera

(2001)concludes that to introduce the technological transformation organizational

culture should be considered. Organizational culture positively considers the envi-

ronment so far, and it also assists the changes (Gordon, 1991). When culture is

ignored and supportive approaches are lacking, the digital transformation of the

supply chain will fail. Organizations face the challenge of taking a step on digital

transformation when they fail to encourage their employees and managers

(Garcia-lorenzo, 2020). So a supportive culture should be maintained or changes

should be made if it is required in the transformation of supply chains.

24 S. Akter et al.

4.3 Blockchain and Supply Chain Management

Nowadays, different supply chain issues are solved through the adoption of

blockchain, such as smart contracts, traceability, product fraud detection, and trust

building (Howson, 2020; Giovanni, 2020; Sunny et al., 2020). Figure 3clariﬁes it

more speciﬁcally. Blockchain has an impact on the traditional supply chain. In this

context, blockchain on SCM is explained in the following section.

Traditional Supply Chain with Blockchain Technology

Traditional SCM has some common strategic objectives (B. Wang et al., 2020). BCT

provides essential assistance to accomplish these objectives efﬁciently and effec-

tively (Kshetri, 2018).

•Cost reduction: Transaction made through BTC minimize the cost by creating an

exclusive code for all transactions. This helps to thoroughly examine the ﬂow of

funds throughout the supply chain discipline process.

•Operational speed: BTC can speed up processing by reducing physical intercon-

nection and transmission.

•Sustainability: BTC can support developing meaningful and computable perfor-

mance metrics to achieve environmental, economic, and social sustainability.

•Risk management: Transactions can only be made when relevant parties agree to

transactions by negotiating among themselves within the blockchain network.

This process supports controlling the data risk of all supply chain transactions

through BCT.

Fig. 3 Implementing Blockchain in Supply Chain Management

Digital Transformation of Supply Chain with Supportive Culture... 25

Fig. 4 Blockchain technology and Supply Chain Management

•Flexibility: BCT can assist customers to locate and track orders from upstream to

downstream, allowing customers to easily change, and also the suppliers to adapt

to instant changes.Basic Characteristics of Blockchain

The characteristics of BCT are explained in this part. BCT establishes visibility,

conﬁdence, order, lucidity, and computerization in a disordered environment

(Viriyasitavat & Hoonsopon, 2018). Blockchain ensures better visibility and security

than traditional supply chain processes. BCT stores speciﬁc information on each

component and provides it to the individual producer in the manufacturing operation

both upstream and downstream (Leary, 2017). Blockchain can be used as an

alternative to improving and replacing paper tracing, speeding up data sharing

(Brent et al., 2013). These data sharing facilities of BCT strengthen the total capacity

to manage the supply chain activities.

Additionally, blockchain keeps a record of business information in a permanent,

veriﬁable, and safe form and keeps track of ownership. That helps the organization

minimize the risk of cybercrime, fraud, and hacking. BTC builds hope among

participants by committing that each record is noted and saved in numerous locations

beyond the whole distributed network. It also increases the skills of supply chain and

reduces the difﬁculty of the system. BTC allows manufacturers and resellers to gain

insight into consumer needs and tailor their products and services in view of that

(Adams et al., 2017).

Transparency/Visibility and Traceability

In a blockchain environment, traceability is deﬁned as the ability to trace and track

data (Sunny et al., 2020). Uses of traceability in the supply chain enhance transpar-

ency. Although traceability and transparency are two interconnected features of BCT

(Wang et al., 2018). The visibility of the supply chain depends a lot on transparency

(Hernandez, 2003). Blockchain ensures better transparency by providing all details

regarding transactions among all parties involved in the supply chain process (Yasin

et al., 2019). Blockchain has a great impact on SCM in traceability and transparency

dimensions (Fig. 4).

26 S. Akter et al.

“Transparency of supply chain is the area in which all its stakeholders have a

shared understanding of, as well as access to, the product-related information that

they desire, without delay, noise, loss, and distortion”(Holland et al., 2017). Product

tracking continues from stats to end, whereas tracing generally towards the origin

from the endpoint. Customers easily gather information about the material, source,

and environmental impact of the product. Manufacturers and distributors beneﬁted

by providing new information to the customer and better product tracking.

There are some main areas in transparency/visibility and traceability. The fol-

lowing are quoted:

•Track the origin of the product.

•Fraud prevention beyond the supply chain network.

•Ensure data security.

Security

The blockchain uses public keys to enhance security and prevent maliciously. The

supply chains of dangerous products should be handled in a very secure manner.

Transforming dangerous goods requires advanced care (Berdik et al., 2021). All

stakeholders involved in the process of hazardous products ﬁnd the appropriate

information through BCT. Producers make smart contact to transport products

with initial information. All parties involved, including the administrative body,

can access this information (Thakur & Breslin, 2020). In this way, the blockchain

creates security through transparency in the supply chain. BCT is built with secure,

“blocks”that store copies of the documents and are oriented to the previous blocks.

This makes them secure and challenging to falsify (Bhushan et al., 2020).

Smart Contracts

Since blockchain is viewed as a more inherently secure form of technology, there is

still a vital role to play for automation. Smart contracts are deﬁned as self-operating

and enhancement applications that use software code and a computing framework to

activate a speciﬁc contract or terms of agreement (Hewa et al., 2020). Smart contract

considers as a complement the use of Distribution Ledger Technology (DLT) and a

decentralized program in the BC network (Han et al., 2020). It can be executed

autonomously in predetermined contexts. The main function of smart contracts is to

implement a peer-to-peer approach without central third-party involvement

(Hu et al., 2021). There is no central dependence on the availability of services in

this system.

5 Findings

This empirical study helps enrich the extant literature on SCM, BCT, supportive

culture, and DT. The present study improves the understanding of how supportive

culture affects supply chain performance in digital transformation. In order to

improve SCM performance in numerous aspects, supportive culture and blockchain

with smart controls play a vital role is identiﬁed in this paper.

Digital Transformation of Supply Chain with Supportive Culture... 27

Those aspects are quoted below:

•Enhancing transparency and traceability helps build a better relationship.

•Reducing the bullwhip effect by providing symmetric information among

partners.

•Detecting fraudulent entries helps to prevent fraud.

•Using smart contracts helps reduce transaction cost and save time.

•By developing a better relationship, providing effective information and

preventing fraud, it creates trust and collaboration among partners.

6 Discussion

Wang et al. (2019) conclude some probable beneﬁts to implementing blockchain in

the supply chain sector, such as increased operational efﬁciency and supply chain

transparency, building mutual trust, and sharing reliable information. The ﬁnding of

Sahebi et al. (2020) indicates that lack of knowledge, cost of employee training, and

vagueness of regulations are the most signiﬁcant barriers to adopting blockchain.

Schmidt and Wagner (2019) concluded that blockchain minimizes operating and

governance cost by automating buyer and supplier contracts and a permanent ledger

of records. The results of Saurabh and Dey (2020) study noted that traceability,

price, consent, faith, dis-intermediation, control, and coordination are the inﬂuential

supply chain actors for implementing BCT.

Liu et al. (2010) found that the organizational culture has diverse effects on the

dimensions of institutional pressures and inter-organizational technological adoption

intention. Sok et al. (2014)ﬁnd that a favorable culture explains the majority of

variance in positive work-to-home meddling and strain-based negative work-to-

home meddling. Blockchain, the internet of things, and AI have the potential to

enrich transparency, faith, and provide substantial assistance by changing national

and organizational culture (Kimani et al., 2020).

Lin (2013)shows that the implementation of e-SCM relies on higher authority

support, absorptive capacity, and competitive pressure. Kshetri (2018) claims that

the supply chain sector is one of the most likely sectors to be transformed into

blockchain. The interconnection between relationship performance and digital trans-

formation is fully mediated by smart technologies (Nasiri et al., 2020). Jabbar et al.

(2020) imply that applying big data is good for every phase of the triple bottom line

in the supply chain. Song et al. (2021)conclude that the introduction of ICT can be

both a warning and an avenue for the wholesale market. Furthermore, marketing

channels and transaction expenses can reduce the attraction of physical wholesale

markets to customers and wholesalers.

28 S. Akter et al.

7 Conclusion, Theoretical Contribution, Policy

Implications, and Future Work

7.1 Conclusion

Today’s world is changing a lot by technological innovation. That is why the online-

based or automated business has taken place rapidly instead of a traditional business.

Organizations implement different modern technologies for different reasons, such

as meeting customer demand, competitor pressure, and the wide acceptance of

technology. Typically, a supportive culture seeks to use the ﬂexibility of the oper-

ating system to link up the needs of employees, maintain interpersonal relationships,

and care for people, thus representing and defending its fundamental beliefs (Sok

et al., 2014). For any kind of organizational change, organizational culture is crucial.

To implement digital transformation, it is necessary to change strategy, leadership,

and organizational culture. The day-by-day organizational culture becomes the basis

of digital transformation in the organization.

DT and analytical methods and novel tactics including DSC can illustrate how to

use different innovative technologies (IoT, cloud computing) to manage supply

chain processes. Blockchain technology is an indicator of digital transformation.

In reducing cost and increasing supply chain performance, BCT plays the driving

role. Most importantly, practicing BCT is more secure, so that only allowed users

can get access the information. That indicates that in facilitating the performance of

SCM, adoption of digital technology more speciﬁcally, BCT is important.

7.2 Theoretical Contribution

This study has an important contribution to supply chain management and organi-

zational supportive culture domains. This paper determines how supportive culture

impacts the adoption of modern innovations such as blockchain technology in SCM.

7.3 Policy Implications

It appears that the ﬁndings should have important implications. Supportive culture is

essential for effective transformation. This paper suggests that stakeholders,

policymakers, and supply chain managers consider organizational culture while

adopting innovative technology. For this reason, the organization has to gain a

deep understanding of cultural complexities and transformation barriers. If an

organization improves its understanding of the relationship among supportive cul-

tural effects, blockchain adoption and the performance of the supply chain will play

an important role in various ﬁelds.

Digital Transformation of Supply Chain with Supportive Culture... 29

7.4 Future Work

However, blockchain technology in SCM is currently in its early stages, and further

studies are needed to extend the present study. Although BCT is becoming a more

widely accepted and recognized topic, there are still many ideas that require future

exploration and analysis. Which can be developed through further research that are

quoted below:

•The relationship among supportive culture, blockchain, and supply chain perfor-

mance in various areas.

•Future investigation is required to develop trust among parties involved in the

supply chain through BCT.

•The blocks in the area of transformation of SCM in blockchain.

•Identify how cultural elements affect supply chain activities to adapt with new

technological changes.

•The operation of smart contracts in SCM should be addressed more in

future work.

This work informs academicians that in the near future, the application of

blockchain in supply chain management will be a new avenue for investigation. It

will be sensational to see what happens over the next decade.

References

Abedin, M. Z., Chi, G., Uddin, M. M., Shahriare, S., Imran, K., & Petr, H. (2020). Tax default

prediction using feature transformation-based machine learning. IEEE Access, 9, 19864–19881.

https://doi.org/10.1109/ACCESS.2020.3048018

Abedin, M. Z., Hassan, M. K., Hajek, P., & Uddin, M. M. (Eds.). (2021). The essentials of machine

learning in ﬁnance and accounting. Routledge.

Adams, R., Parry, G., Godsiff, P., & Ward, P. (2017). The future of money and further applications

of the blockchain. Strategic Change, 26(5), 417–422. https://doi.org/10.1002/jsc.2141

Agrawal, P., Narain, R., & Ullah, I. (2020). Analysis of barriers in implementation of digital

transformation of supply chain using interpretive structural modelling approach. Journal of

Modelling in Management, 15(1), 297–317. https://doi.org/10.1108/JM2-03-2019-0066

Ali, O., Ally, M., & Dwivedi, Y. (2020). The state of play of blockchain technology in the ﬁnancial

services sector: A systematic literature review. International Journal of Information Manage-

ment, 54, 102199. https://doi.org/10.1016/j.ijinfomgt.2020.102199

APQC. (2018). Quick poll: Current state of robotic process automation in the supply chain. Supply

Chain Management Review. Accessed from https://www.scmr.com/article/quick_poll_current_

state_of_robotic_process_automation_in_the_supply_chain

Aste, T., & Matteo, T. D. (2017). Blockchain technologies: The foreseeable impact on society and

industry. Computer, 50(9), 18–28. https://doi.org/10.1109/MC.2017.3571064

Azzi, R., Kilany, R., & Sokhn, M. (2019). The power of a blockchain-based supply chain.

Computers & Industrial Engineering, 135, 582–592. https://doi.org/10.1016/j.cie.2019.06.042

Banks, G. C., Dionne, S. D., Sayama, H., Schmid, M., & Special, M. (2019). Leadership in the

digital era: Social media, big data, virtual reality, computational methods, and deep learning.

The Leadership Quarterly, 30(3), 3–4. https://doi.org/10.1016/S1048-9843(19)30294-2

30 S. Akter et al.

Berdik, D., Otoum, S., Schmidt, N., Porter, D., & Jararweh, Y. (2021). A survey on Blockchain for

information systems management and security. Information Processing and Management,

58(1), 102397. https://doi.org/10.1016/j.ipm.2020.102397

Bhushan, B., Sinha, P., Sagayam, K. M., & Andrew, J. (2020). Untangling blockchain technology:

A survey on state of the art, security threats, privacy services, applications and future research

directions. Computers and Electrical Engineering, 90, 106897. https://doi.org/10.1016/j.

compeleceng.2020.106897

Bicocchi, N., Cabri, G., Mandreoli, F., & Mecella, M. (2019). Dynamic digital factories for agile

supply chains: An architectural approach. Journal of Industrial Information Integration, 15,

111–121. https://doi.org/10.1016/j.jii.2019.02.001

Bollinger, A. S., Smith, R. D., Bollinger, A. S., & Smith, R. D. (2002). Managing organizational

knowledge as a strategic asset. Journal of Knowledge Management, 5(1), 8–18. https://doi.org/

10.1108/13673270110384365

Bounfour, A. (2016). Digital futures, digital transformation: From lean production to acceluction.

Springer.

Brent, A., Joseph, D. W., Travis, R., & Morgan, T. (2013). Leveraging supply chain visibility for

responsiveness: The moderating role of internal integration. Journal of Operations Manage-

ment, 31(7–8), 543–554. https://doi.org/10.1016/j.jom.2013.09.003

Büyüközkan, G., & Göçer, F. (2018). Computers in industry digital supply chain: Literature review

and a proposed framework for future research. Computers in Industry, 97, 157–177. https://doi.

org/10.1016/j.compind.2018.02.010

Cabrera, E. F. (2001). The key role of organizational culture in a multi-system view of technology-

driven change. International Journal of Information Management, 21(3), 245–261.

Carbonneau, R., Laframboise, K., & Vahidov, R. (2008). Application of machine learning tech-

niques for supply chain demand forecasting. European Journal of Operational Research, 184,

1140–1154. https://doi.org/10.1016/j.ejor.2006.12.004

Chen, Y. (2018). Blockchain tokens and the potential democratization of entrepreneurship and

innovation. Business Horizons, 61(4), 567–575. https://doi.org/10.1016/j.bushor.2018.03.006

Choi, T., Feng, L., & Li, R. (2020). Information disclosure structure in supply chains with rental

service platforms in the Blockchain technology era. International Journal of Production

Economics, 221, 107473. https://doi.org/10.1016/j.ijpe.2019.08.008

Da Xu, L., Xu, E. L., & Li, L. (2018). Industry 4.0: State of the art and future trends. International

Journal of Production Research, 56(8), 2941–2962. https://doi.org/10.1080/00207543.2018.

1444806

Di, D., Maesa, F., & Mori, P. (2020). Blockchain 3.0 applications survey. Journal of Parallel and

Distributed Computing, 138,99–114. https://doi.org/10.1016/j.jpdc.2019.12.019

Dowty, R. A., & Wallace, W. A. (2010). Implications of organizational culture for supply chain

disruption and restoration. International Journal of Production Economics, 126(1), 57–65.

https://doi.org/10.1016/j.ijpe.2009.10.024

Farhana, A., Petr, H., Munk, M., Abedin, M. Z., Shahriare, S., Efat, M. I. A., & Islam, J. (2022). A

systematic review of blockchain applications. IEEE Access.

Garcia-lorenzo, L. (2020). Organizational remembering as a trigger for cultural change: Exploring

the episodic memories of a ﬁnancial scandal. Scandinavian Journal of Management, 36(1),

101089. https://doi.org/10.1016/j.scaman.2019.101089

Ghasemian, I., Masoomi, B., & Ghorbani, S. (2020). Technology in society expert oriented

approach for analyzing the blockchain adoption barriers in humanitarian supply chain. Tech-

nology in Society, 63, 101427. https://doi.org/10.1016/j.techsoc.2020.101427

Giovanni, P. D. (2020). Blockchain and smart contracts in supply chain management: A game

theoretic model. International Journal of Production Economics, 228, 107855. https://doi.org/

10.1016/j.ijpe.2020.107855

Gordon, G. G. (1991). Industry determinants of organizational culture. Academy of Management

Review, 16(2), 396–415.

Digital Transformation of Supply Chain with Supportive Culture... 31

Halpern, N., Mwesiumo, D., Suau-sanchez, P., Budd, T., & Bråthen, S. (2021). Ready for digital

transformation? The effect of organisational readiness, innovation, airport size and ownership

on digital change at airports. Journal of Air Transport Management, 90, 101949. https://doi.org/

10.1016/j.jairtraman.2020.101949

Han, D., Zhang, C., Ping, J., & Yan, Z. (2020). Smart contract architecture for decentralized energy

trading and management based on blockchains. Energy, 199, 117417. https://doi.org/10.1016/j.

energy.2020.117417

Hartley, J. L., & Sawaya, W. J. (2019). Tortoise, not the hare: Digital transformation of supply chain

business processes. Business Horizons, 62(6), 707–715. https://doi.org/10.1016/j.bushor.2019.

07.006

Hernandez, F. G. H. (2003). Derivatives and the FASB: Visibility and transparency? Critical

Perspectives on Accounting, 14(8), 777–789. https://doi.org/10.1016/S1045-2354(02)00192-2

Hewa, T., Ylianttila, M., & Liyanage, M. (2020). Survey on blockchain based smart contracts:

Applications, opportunities and challenges. Journal of Network and Computer Applications,

177, 102857. https://doi.org/10.1016/j.jnca.2020.102857

Holland, D., Krause, A., Provencher, J., & Seltzer, T. (2017). Transparency tested: The inﬂuence of

message features on public perceptions of organizational transparency. Public Relations Review,

44(2), 256–264. https://doi.org/10.1016/j.pubrev.2017.12.002

Howson, P. (2020). Building trust and equity in marine conservation and ﬁsheries supply chain

management with blockchain. Marine Policy, 115, 103873. https://doi.org/10.1016/j.marpol.

2020.103873

Hu, T., Liu, X., Chen, T., Zhang, X., & Huang, X. (2021). Transaction-based classiﬁcation and

detection approach for Ethereum smart contract. Information Processing & Management, 58(2),

102462.

Huang, F., & Vasarhelyi, M. A. (2019). Applying robotic process automation (RPA) in auditing: A

framework. International Journal of Accounting Information Systems, 35, 100433. https://doi.

org/10.1016/j.accinf.2019.100433

Isensee, C., Teuteberg, F., Griese, K., & Topi, C. (2020). The relationship between organizational

culture, sustainability, and digitalization in SMEs: A systematic review. Journal of Cleaner

Production, 275, 122944. https://doi.org/10.1016/j.jclepro.2020.122944

Jabbar, S., Lloyd, H., Hammoudeh, M., Adebisi, B., & Raza, U. (2020). Blockchain-enabled supply

chain: Analysis, challenges, and future directions. Multimedia Systems, 27, 787–806. https://doi.

org/10.1007/s00530-020-00687-0

Jabbour, C. J. C., Fiorini, P. D. C., Ndubisi, N. O., Queiroz, M. M., & Piato, É. L. (2020). Digitally-

enabled sustainable supply chains in the 21st century: A review and a research agenda. Science

of the Total Environment, 725, 138177. https://doi.org/10.1016/j.scitotenv.2020.138177

Kar, A. K., & Navin, L. (2021). Telematics and informatics diffusion of blockchain in insurance

industry: An analysis through the review of academic and trade literature. Telematics and

Informatics, 58, 101532. https://doi.org/10.1016/j.tele.2020.101532

Karamchandani, A., Srivastava, S. K., & Srivastava, R. K. (2019). Perception-based model for

analyzing the impact of enterprise blockchain adoption on SCM in the Indian service industry.

International Journal of Information Management, 52, 102019. https://doi.org/10.1016/j.

ijinfomgt.2019.10.004

Karine, C. (2020). The relationship between presenteeism and job satisfaction: A mediated mod-

eration model using work engagement and perceived organizational support. European Man-

agement Journal, 39(2), 270–278. https://doi.org/10.1016/j.emj.2020.09.001

Khan, M., Nasﬁkur, R., Bala, S., Yesmin, S., & Abedin, M. Z. (2022). Bioinformatics: The

importance of data mining techniques. In Sentimental analysis and deep learning

(pp. 411–422). Springer.

Kimani, D., Adams, K., Attah-Boakye, R., Ullah, S., Frecknall-Hughes, J., & Kim, J. (2020).

Blockchain, business and the fourth industrial revolution: Whence, whither, wherefore and

how? Technological Forecasting and Social Change, 161, 120254.

Kokina, J., & Blanchette, S. (2019). Early evidence of digital labor in accounting: Innovation with

robotic process automation. International Journal of Accounting Information Systems, 35,

100431. https://doi.org/10.1016/j.accinf.2019.100431

32 S. Akter et al.

Kshetri, N. (2018). 1 Blockchain’s roles in meeting key supply chain management objectives.

International Journal of Information Management, 39,80–89. https://doi.org/10.1016/j.

ijinfomgt.2017.12.005

Lafourcade, P., & Lombard-platet, M. (2020). About blockchain interoperability. Information

Processing Letters, 1, 105976. https://doi.org/10.1016/j.ipl.2020.105976

Leary, D. E. O. (2017). Conﬁguring blockchain architectures for transaction information in

blockchain consortiums: The case of accounting and supply chain systems. Intelligent Systems

in Accounting, Finance and Management, 24(4), 138–147. https://doi.org/10.1002/isaf.1417

Li, W., Liu, K., Belitski, M., Ghobadian, A., & Regan, N. O. (2016). Sized enterprises in the digital

age. Journal of Information Technology, 31(2), 185–206. https://doi.org/10.1057/jit.2016.10

Lin, H. (2013). Understanding the determinants of electronic supply chain management system

adoption: Using the technology–organization–environment framework. Technological Fore-

casting and Social Change, 86,80–92. https://doi.org/10.1016/j.techfore.2013.09.001

Liou, D., Tu, C., & Chang, S. (2012). Mediating effect between supportive culture and job

satisfaction in administrative services at higher education institutions. Human Factors and

Ergonomics in Manufacturing & Service Industries, 24(6), 627–640.

Liu, H., Ke, W., Kee, K., Gu, J., & Chen, H. (2010). The role of institutional pressures and

organizational culture in the ﬁrm’intention to adopt internet-enabled supply chain management

systems. Journal of Operations Management, 28(5), 372–384. https://doi.org/10.1016/j.jom.

2009.11.010

Michela, J. L., & Burke, W. W. (2000). Organizational culture and climate in transformations for

quality and innovation. In Handbook of organizational culture and climate (pp. 225–244).

Springer.

Muthukrishnan, N., Maleki, F., & Ovens, K. (2020). Brief history of artiﬁcial intelligence. Neuro-

imaging Clinics, 30(4), 393–399. https://doi.org/10.1016/j.nic.2020.07.004

Nakamoto, S. (2008). Bitcoin: A peer-to-peer electronic cash system. Decentralized Business

Review, 21260,1–9.

Nam, K., Dutt, C. S., Chathoth, P., Khan, M. S., & Dutt, C. S. (2021). Blockchain technology for

smart city and smart tourism: Latest trends and challenges. Asia Paciﬁc Journal of Tourism

Research, 26, 454–468. https://doi.org/10.1080/10941665.2019.1585376

Nasiri, M., Ukko, J., Saunila, M., & Rantala, T. (2020). Technovation managing the digital supply

chain: The role of smart technologies. Technovation, 96–97, 102121. https://doi.org/10.1016/j.

technovation.2020.102121

Papagiannidis, S., Harris, J., & Morton, D. (2020). WHO led the digital transformation of your

company? A reﬂection of IT related challenges during the pandemic. International Journal of

Information Management, 55, 102166.

Petr, H., & Abedin, M. Z. (2020). A proﬁt function-maximizing inventory backorder prediction

system using big data analytics. IEEE Access, 8, 58982–58994. https://doi.org/10.1109/

ACCESS.2020.2983118

Prasitlumkum, N., Cheungpasitporn, W., Chokesuwattanaskul, A., Thangjui, S., Thongprayoon, C.,

Bathini, T., Vallabhajosyula, S., Kanitsoraphan, C., Leesutipornchai, T., &

Chokesuwattanaskul, R. (2020). Diagnostic accuracy of smart gadgets/wearable devices in

detecting atrial ﬁbrillation: A systematic review and meta-analysis. Archives of Cardiovascular

Diseases, 114(1), 4–16. https://doi.org/10.1016/j.acvd.2020.05.015

Queiroz, M. M., & Fosso, S. (2019). Blockchain adoption challenges in supply chain: An empirical

investigation of the main drivers in India and the USA. International Journal of Information

Management, 46,70–82. https://doi.org/10.1016/j.ijinfomgt.2018.11.021

Sahebi, I. G., Masoomi, B., & Ghorbani, S. (2020). Expert oriented approach for analyzing the

blockchain adoption barriers in humanitarian supply chain. Technology in Society, 63, 101427.

Sana, J. K., Abedin, M. Z., Rahman, M. S., & Rahman, M. S. (2022). Data transformation based

optimized customer churn prediction model for the telecommunication industry. arXiv preprint

arXiv:2201.04088.

Saurabh, S., & Dey, K. (2020). Blockchain technology adoption, architecture, and sustainable agri-

food supply chains. Journal of Cleaner Production, 284, 124731. https://doi.org/10.1016/j.

jclepro.2020.124731

Digital Transformation of Supply Chain with Supportive Culture... 33

Schmidt, C. G., & Wagner, S. M. (2019). Blockchain and supply chain relations: A transaction cost

theory perspective. Journal of Purchasing and Supply Management, 25(4), 100552. https://doi.

org/10.1016/j.pursup.2019.100552

Shajalal, M., Hajek, P., & Abedin, M. Z. (2021). Product backorder prediction using deep neural

network on imbalanced data. International Journal of Production Research, 1–18,1.https://doi.

org/10.1080/00207543.2021.1901153

Shi, S., He, D., Li, L., & Kumar, N. (2020). Applications of blockchain in ensuring the security and

privacy of electronic health record systems: A survey. Computers & Security, 97, 101966.

https://doi.org/10.1016/j.cose.2020.101966

Singh, M., & Kim, S. (2018). Branch based blockchain technology in intelligent vehicle. Computer

Networks, 145, 219–231. https://doi.org/10.1016/j.comnet.2018.08.016

Sok, J., Blomme, R., & Tromp, D. (2014). Positive and negative spillover from work to home: The

role of organizational culture and supportive arrangements. British Journal of Management,

25(3), 456–472. https://doi.org/10.1111/1467-8551.12058

Song, Y., Escobar, O., Arzubiaga, U., & Massis, A. D. (2021). The digital transformation of a

traditional market into an entrepreneurial ecosystem. Review of Managerial Science, 16(1),

65–88. https://doi.org/10.1007/s11846-020-00438-5

Sunny, J., Undralla, N., & Pillai, V. M. (2020). Supply chain transparency through blockchain-

based traceability: An overview with demonstration. Computers & Industrial Engineering, 150,

106895. https://doi.org/10.1016/j.cie.2020.106895

Thakur, S., & Breslin, J. G. (2020). Scalable and secure product serialization for multi-party

perishable good supply chains using blockchain. Internet of Things, 11, 100253. https://doi.

org/10.1016/j.iot.2020.100253

Toorajipour, R., Sohrabpour, V., Nazarpour, A., Oghazi, P., & Fischl, M. (2021). Artiﬁcial

intelligence in supply chain management: A systematic literature review. Journal of Business

Research, 122, 502–517. https://doi.org/10.1016/j.jbusres.2020.09.009

Verhoef, P. C., Broekhuizen, T., Bart, Y., Bhattacharya, A., Qi, J., Fabian, N., & Haenlein,

M. (2019). Digital transformation: A multidisciplinary reﬂection and research agenda. Journal

of Business Research, 122, 889–901. https://doi.org/10.1016/j.jbusres.2019.09.022

Viriyasitavat, W., & Hoonsopon, D. (2018). Cultivating product innovation performance through

creativity: The impact of organizational agility and ﬂexibility under technological turbulence.

Journal of Industrial Information Integration, 33(4), 741–762. https://doi.org/10.1016/j.jii.

2018.07.004

Wang, X., Fu, D., Fruk, G., Chen, E., & Zhang, X. (2018). Improving quality control and

transparency in honey peach export chain by a multi-sensors-managed traceability system.

Food Control, 88, 169–180. https://doi.org/10.1016/j.foodcont.2018.01.008

Wang, Y., Singgih, M., Wang, J., & Rit, M. (2019). Title page making sense of blockchain

technology: How will it transform supply chains? International Journal of Production Eco-

nomics, 211, 221–236. https://doi.org/10.1016/j.ijpe.2019.02.002

Wang, B., Luo, W., Zhang, A., Tian, Z., & Li, Z. (2020). Computers in industry blockchain-enabled

circular supply chain management: A system architecture for fast fashion. Computers in

Industry, 123,103324. https://doi.org/10.1016/j.compind.2020.103324

Wu, J. Y. (2008). A general behavior model and new deﬁnitions of organizational cultures. The

Journal of Socio-Economics, 37(6), 2535–2545. https://doi.org/10.1016/j.socec.2007.10.002

Xuan, S., Zheng, L., Chung, I., Wang, W., Man, D., Du, X., Yang, W., & Guizani, M. (2020). An

incentive mechanism for data sharing based on blockchain with smart contracts R. Computers

and Electrical Engineering, 83, 106587. https://doi.org/10.1016/j.compeleceng.2020.106587

Yang, F., Qiao, Y., Abedin, M. Z., & Huang, C. (2022). Privacy-preserved credit data sharing

integrating Blockchain and federated learning for industrial 4.0. IEEE Transactions on Indus-

trial Informatics, 2022,1.https://doi.org/10.1109/TII.2022.3151917

Yasin, M., Sabir, M., & Ali, H. (2019). CertLedger: A new PKI model with certiﬁcate transparency

based on blockchain. Computers & Security, 85, 333–352. https://doi.org/10.1016/j.cose.2019.

05.013

Integration of Artiﬁcial Intelligence

Technology in Management Accounting

Information System: An Empirical Study

Emon Kalyan Chowdhury

Abstract At present, most of the business organizations take their management

decisions using traditional approach. In the traditional approach, the freedom to be

ﬂexible is limited due to numerous assumptions. This paper aims to establish an

artiﬁcial neural network-based model to predict management information and verify

the accuracy of the model using some real data. The proposed model covers ﬁve

dimensions, namely, accounting analysis management system, accounting decision

support system, performance management information system, risk management

information system, and environmental management information system. It is

observed that the proposed model can predict the management accounting informa-

tion by 98.83%, which is extremely good and meets the accounting information

requirement.

Keywords Artiﬁcial intelligence · Machine learning · Management accounting ·

Information system · Neural network

1 Introduction

Management accounting provides information to managers who make important

decisions in an organization (Garrison et al., 2003). The size and complexity of data

is increasing day by day as a result managers are in serious trouble in processing

large amount of data (Munim et al., 2020). The success of a decision depends on the

quality of the information. Therefore, an efﬁcient management accounting informa-

tion system where data are processed through artiﬁcial intelligence technology plays

a vital role in improving the operating efﬁciency of an organization (Zhang, 2021).

Management enterprises are substantially dependent on advanced information

technology to make rational and effective decisions. Among management informa-

tion systems, the management accounting information system is the most important

E. K. Chowdhury (✉)

CIU Business School, Chittagong Independent University, Chattogram, Bangladesh

©The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

M. Z. Abedin, P. Hajek (eds.), Novel Financial Applications of Machine Learning

and Deep Learning, International Series in Operations Research & Management

Science 336, https://doi.org/10.1007/978-3-031-18552-6_3

segment (Hutahayan, 2020). The signiﬁcance of management accounting informa-

tion system lies in the economic progress, expansion, scale of economies, acquisi-

tion, and continuous improvement of strengths through scientiﬁc management

decisions (Cai et al., 2019).

36 E. K. Chowdhury

Practically, the use of management accounting information system is conﬁned to

the cost management, preparation of different budgets, and performance manage-

ment. Smooth functioning of enterprise management is highly dependent on the

comprehensive and stable construction of management accounting information

systems integrated with other management information systems (Goetz et al., 2015).

The remaining part continues as follows. Section 2reviews previous studies.

Section 3analyzes different models based on artiﬁcial intelligence technology.

Section 4experiments the success rates of prediction capacity of model using

authentic management information data, and Sect. 5concludes the paper.

2 Literature Review

Management control systems ensure optimal use of limited resources to achieve the

organization’s goal. In addition to ﬁnancial data, an efﬁcient management control

system also uses psychological and control variables (Nguyen et al., 2017). The data

from multiple sources are collected and fed into the management information system

so as to generate various sub-objectives from a single organizational objective. It

helps to compare the actual performance with the projected plans from diverse

perspectives (Al-Ali et al., 2017). To sustain itself in a competitive and

technology-based environment, an organization must strengthen its managerial and

supervisory functions by introducing a management control system (Chi et al., 2019;

Xin et al., 2018). Out of the different wings of the management information system,

the development of the management accounting information system is crucial, as it

directly contributes to the organization’sﬁnancial solvency, internal control system,

customer retention, and overall sustainability (Chowdhury, 2019; Ward et al., 2016).

Recently, the use of an e-commerce-based accounting information system has

increased tremendously among the enterprises to enjoy competitive advantages

(Shajalal et al., 2021; Hidayat et al., 2020). Management accounting plays an

important role in fulﬁlling the economic needs of an organization’s operation and

management with the help of responsibility center. The responsibility center ensures

optimum uses of internal accounting control systems and further assists in organiz-

ing and delivering other functional internal management systems (Ghasemi et al.,

2019). Amershi et al. (2014) observed a signiﬁcantly positive impact of management

accounting on innovation management. Management accounting systems simplify

the cost calculation of single and batch products (Rodriguez-Galiano et al., 2015).

Cooper et al. (2017) noticed the increasing popularity of using balanced scorecards

in organizations to measure the performance of different indicators.

The traditional management accounting system mostly depends on the assump-

tions rather than versatility of data, which imperatively directs to take ﬁxed

decisions. This study ﬁnds a gap to explore the possibility of taking dynamic

decisions by using alternative models where artiﬁcial intelligence technology is

used in line with machine learning and data mining algorithms.

Integration of Artiﬁcial Intelligence Technology in Management... 37

3 Artiﬁcial Neural Network (ANN)

The design of ANN is inspired by the structure of biological neurons such as the

human brain. In a human brain, neurons create a network through interconnections.

A neuron is known as a cell and executes a single task by responding to an input

signal. In an ANN, the nodes are connected to each other and establish a network

among themselves. The nodes are designed using artiﬁcial intelligence to handle

massive amount of data using multiple equations simultaneously. In this network,

the equations are established through sequential computations following a trial-and-

error approach (Abedin et al., 2021; Chakraborty et al., 2018). The basic structure of

ANN is expressed in Fig. 1.

Input neurons X

,...,X

indicate various inputs to the network, synapse

weights W

,...,W

signify the weights of connections. The weights are very

important in ANN as these represent the strength of each node. The weights that

govern the effect of neurons are measured in the numerical parameters, which

determines the output by converting the input.

The hidden layer performs the processing task. It applies two operational func-

tions, the summation function and the transfer or activation function. The summation

function multiplies each input (X

) with the corresponding weight (W

) and all

products (W

×X

) result in the summation function ξ=∑W

×X

+B, where

Brepresents the bias value. It controls the output of the neuron in line with the

weighted sum of inputs.

The activation function transforms the input signal from the summation function

into to output of a node for an ANN model. Each ANN is made up of three

components. First, the node character determines inputs and outputs through signal

processing. Second, the network topology determines how the nodes are connected

summation

function

ξ=∑W

transfer/activa-

tion function

Y=∫(ξ)

Bias

…

Fig. 1 Model of an artiﬁcial neuron

and organized. Third, the learning rules create and adjust weights. A few widely used

ANN-based models have been discussed below.

38 E. K. Chowdhury

Offer

charge

cards

Offer

credit

cards

Offer

cards

Have a

worldwide

presence

Offer jointly

branded

international

currency cards

Provide

Range of

Services

Yearly

sales

Offer

internation

al banking

Increase

revenue

Acquire

other

companies

Weaker

US Dollar

Increase

Sales

Decrease

costs

Yearly

costs

Collect

Interest

Collect

Subscription

fee

Select

type of

card(s)

Facilitate

card

processing

Strong

economic

growth

Stay

competitive

Accurate

transactions

#Data

entry

errors

Handle

transaction

volumes

Minimize

international

conversion

costs

International

conversion

costs

Credit card

transaction

Make

agreements

with other credit

card companies

Translate

revenue and

costs across

currencies

State-of-the-

art transaction

systems

International

Development

International

development

program

Legend

Goal

Indicator

Refinemen t

Influence

AND

Task

Situation

(External)

Evaluates

Measures

Situation

(Internal)

Fig. 2 Business intelligence model for a credit card company. Source: Horkoff et al. (2012)

Business Intelligence (BI) Model

BI helps an organization excel at operational activities in such a way that helps tap

the opportunities in the market while overcoming potential threats. It has the

capacity to generate effective information to take strategic decisions by processing

massive volume of data. BI establishes a network between an organization and the

external environment with the support of different reasoning techniques that controls

inﬂuences, situations, and the indicators (Fig. 2). The reasoning tools for this model

are “what if”a bottom-up approach and “is it possible?”a top-down approach

(Horkoff et al., 2012).

Three-Tier Data Model

The three-tier data model is widely used in the data warehouse management of an

organization. It provides subject-wise analytical environment in the global context

(Abedin et al., 2020; Lau et al., 2018). The three tiers have been outlined below:

(a) Conceptual model: This is the top level of the model which is expressed by

topics. Topics are derived from the dimensions and measures. Dimension refers

to a perspective through which people observe the world, and measurement is

related to data information of the dimension. For example, sales volume.

(b) Logical model: Logical model may be classiﬁed into two models such as the star

model and the snowﬂake model. The star model includes the fact table and

dimension model, while at the same time, they are connected to each other. The

Integration of Artiﬁcial Intelligence Technology in Management... 39

Data dimension

Year

Quarter

Month

Week

Day

Customer ID

Name

Address

City

Store dimension

City

State

District

Zip

Product dimension

Name

Description

Price

Brand

Sales Fact

Sales ID

Customer ID

Product ID

Date ID

Store ID

Sales units

Fig. 3 Star model

Data dimension

Year

Quarter

Month

Week

Day

Store dimension

City

State

District

Zip

Product dimension

Name

Description

Price

Brand

Sales Fact

Sales ID

Customer ID

Product ID

Date ID

Store ID

Sales units Product variant

Variant ID

Variant name

Product type

Customer ID

Name

Address

City

Fig. 4 Snowﬂake model

star model is shown in Fig. 3. The sales data are generated in different time

dimension including customer details, store details, and product details.

A snowﬂake model is an extension of the star model. It includes additional

information about a particular dimension (Fig. 4). It uses similar disk space, is

easy to install, and reduces query performance for multiple tables.

Extract, Transform, Load (ETL) Model

In this model, data are extracted from multiple source systems and then converted to

ﬁnal data after necessary calculations. The converted data are loaded into the data

warehouse system for managerial decision. Source points include relevant stake-

holders such as analysts, developers, testers, and top brass executives. Since ETL

activities occur regularly, the data warehouse required to be updated, agile, and

properly documented. ETL helps to make critical business decisions, and compare

the data of the source and target system through data migration and manipulation.

Where the transactional database fails to answer complex business questions, ETL

can easily and quickly address them (Hajek & Abedin, 2020; Sabtu et al., 2017).

Figure 5shows the ETL process in three steps.

40 E. K. Chowdhury

Oracle

SQL

Teradata

Flat file

Staging Area Data Warehouse

Fig. 5 ETL model

In the ETL model, data are fed into the staging area by extracting them from the

source points after due validations. Data are extracted from the source points in raw

format, and at the transformation stage, data are cleaned, mapped, and converted. In

this stage, the ETL assigns values and modiﬁes the data so that business intelligence-

based reports can be generated. Warehousing data is the last step of the ETL model.

Here, a huge volume of data can be loaded in signiﬁcantly less time. If the loading

process fails, the recovery mechanism is activated without failure of any sort of data

integrity. The entire ETL process is controlled by the warehouse administrator

(Abedin et al., 2018).

Cube Structure

The data cube is a three-dimensional way of presenting data. In this model, the data

are judged from various perspectives. When data cannot be presented in traditional

column and row format due to more variables and context, data cube can make it so

simple by utilizing different angles (Augenstein et al., 2018). Data cubes have the

following categories.

(a) Multidimensional data cube: Most of the online analytical processing (OLAP)

products are designed using a multidimensional array. These OLAPs perform

better than other approaches, as they can be indexed straight to collect subsets of

data. The larger the dimension, the sparser the cubes.

(b) Rational OLAP (ROLAP): This model uses a relational database to store and

manage warehouse data. ROLAP servers are highly scalable and analyze

Integration of Artiﬁcial Intelligence Technology in Management... 41

massive volumes of data across multiple dimensions. It also stores and analyzes

highly volatile and changeable data.

To understand the presentation of the data in cube structure, the following

information can be considered (Table 1).

The above information is shown in a three-dimensional cube (Fig. 6).

The essence of the cube structure lies in the capacity to show different data in a

single image.

Data Mining (DM) Process

DM is an essential part of the management accounting information system (Kara

et al., 2020). It combines database, statistics, machine learning, and other relevant

technologies. It generates required information for managers amalgamating different

data to enjoy competitive advantages (Abedin et al., 2019). Figure 7depicts the data

mining process.

4 Proposed Model

In light of the above analysis, this study recommends an Intelligent Management

Accounting Information System (IMAIS) for the decision-making process where the

following aspects are integrated. This model is the extension of Zhang (2021) where

the environmental management information system was not included. In this model,

the impact of the management decision on the environment has been considered. The

integrated systems are as follows:

(a) Accounting analysis management system

(b) Performance management information system

(d) Risk management information system, and

(e) Environmental management information system

This recommended model can provide customized information to take decisions

in time and also helps to run its business in a way better ensuring a sound internal

control system. Figure 8shows an IMAIS formation structure.

The recommended IAMAIS model covers reporting systems, risk management,

performance management, decision support issues, and environmental issues. Each

sub-system works autonomously and combinedly to fulﬁll segment and enterprise

requirements.

Test of Model Efﬁciency

To verify the degree of accuracy of the proposed model, this study has used real

management accounting data. Out of 380 observations, a total of 125 observations

have been used classifying into 13 categories to train the model. The predicted

results and actual results are shown in Fig. 9.

42 E. K. Chowdhury

Table 1 Location-wise quarterly data

Location =“Chicago”Location =“New York”Location =“Toronto”

Item Item Item

Home Home Home

Time Ent. Comp. Phone Sec. Ent. Comp. Phone Sec. Ent. Comp. Phone Sec.

Q1 854 882 89 623 1087 968 38 872 818 746 43 591

Q2 943 890 64 698 1130 1024 41 925 894 769 52 682

Q3 1032 924 59 789 1034 1048 45 1002 940 795 58 728

Q4 1129 992 63 870 1142 1091 54 984 978 864 59 784

Integration of Artiﬁcial Intelligence Technology in Management... 43

Fig. 6 Cube structure

Fig. 7 Data mining process

Validati

Analysis

Pre-

processi

Data

input

Fig. 8 Intelligent

management accounting

information system

Accounting

analysis

management

system

Performance

management

information

system

Accounting

decision support

system

Risk

management

information

system

Environmental

management

information

system

It is observed that the prediction is very close to the actual results for most of the

observations. To get a further clear scenario, the residuals of the actual and predicted

data are shown in Fig. 10.

It is also observed that most residuals hover within 0.05 to -0.05 and a very

insigniﬁcant number of observations are above 0.1 to -0.01. This clearly indicates

that the model is capable of predicting management information with an accuracy

rate of 98.83%. As the rate is very close to 100%, it may be applied in the real world.

44 E. K. Chowdhury

0.2

0.4

0.6

0.8

1.2

0 50 100 150 200 250 300 350 400

Predicted Actual

Fig. 9 Actual vs. predicted data

-0.15

-0.1

-0.05

0.05

0.1

0.15

0.2

0 50 100 150 200 250 300 350 400

Fig. 10 The residuals of actual vs. predicted results

5 Conclusion

This study aimed to formulate a management accounting information system using

machine learning and an artiﬁcial neural network model. Being a vital

sub-information system of management information system, the management

accounting information system plays a very important role in the accounting

development, therefore it should incorporate the accounting analysis management

system, performance management information system, accounting decision support

system, risk management information system, and environmental management

information system. The recommended model can predict the accounting data with

an accuracy rate of 98.83%. As the business world is complex and affected by many

factors, the use of artiﬁcial intelligence technology to make management accounting

decisions knows no bounds. It is assumed that the synergy of ﬁve dimensions helps

in taking appropriate business decisions. Future researchers may include legal and

ethical issues in the model to make this model more reliable and applicable as these

issues vary from country to country.

Integration of Artiﬁcial Intelligence Technology in Management... 45

References

Abedin, M. Z., Chi, G., Colombage, S., & Moula, F. E. (2018). Credit default prediction using a

support vector machine and a probabilistic neural network. Journal of Credit Risk. Accessed

from https://ssrn.com/abstract=3175776

Abedin, M. Z., Guotai, C., Moula, F. E., Azad, A. S., & Khan, M. S. U. (2019). Topological

applications of multilayer perceptrons and support vector machines in ﬁnancial decision support

systems. International Journal of Finance & Economics, 24(1), 474–507.

Abedin, M. Z., Chi, G., Uddin, M. M., Satu, M. S., Khan, M. I., & Hajek, P. (2020). Tax default

prediction using feature transformation-based machine learning. IEEE Access, 9, 19864–19881.

Abedin, M. Z., Hassan, M. K., Khan, I., & Julio, I. F. (2021). Feature transformation for corporate

tax default prediction: Application of machine learning approaches. Asia-Paciﬁc Journal of

Operational Research, 2140017.

Al-Ali, A. R., Zualkernan, I. A., Rashid, M., Gupta, R., & AliKarar, M. (2017). A smart home

energy management system using IoT and big data analytics approach. IEEE Transactions on

Consumer Electronics, 63(4), 426–434.

Amershi, S., Cakmak, M., Knox, W. B., & Kulesza, T. (2014). Power to the people: The role of

humans in interactive machine learning. AI Magazine, 35(4), 105–120.

Augenstein, D., Fleig, C., & Maedche, A. (2018, June). Development of a data-driven business

model transformation tool. In International Conference on Design Science Research in Infor-

mation Systems and Technology (pp. 205–217). Springer.

Cai, J., Huang, W., Yang, S., Wang, S., & Luo, J. (2019, August). A selection method for Denoising

auto encoder features using cross entropy. In International Conference on Intelligent Computing

(pp. 479–490). Springer.

Chakraborty, T., Chattopadhyay, S., & Chakraborty, A. K. (2018). A novel hybridization of

classiﬁcation trees and artiﬁcial neural networks for selection of students in a business school.

Opsearch, 55(2), 434–446.

Chi, G., Uddin, M. S., Abedin, M. Z., & Yuan, K. (2019). Hybrid model for credit risk prediction:

An application of neural network approaches. International Journal on Artiﬁcial Intelligence

Tools, 28(05), 1950017.

Chowdhury, E. K. (2019). Transformation of business model through blockchain technology.

Accounting and Finance, 47(5), 4–9.

Cooper, D. J., Ezzamel, M., & Qu, S. Q. (2017). Popularizing a management accounting idea: The

case of the balanced scorecard. Contemporary Accounting Research, 34(2), 991–1025.

Garrison, R. H., Noreen, E. W., Brewer, P. C., & Mardini, R. U. (2003). Managerial accounting.

McGraw-Hill/Irwin.

Ghasemi, R., Habibi, H. R., Ghasemlo, M., & Karami, M. (2019). The effectiveness of management

accounting systems: Evidence from ﬁnancial organizations in Iran. Journal of Accounting in

Emerging Economies, 9(2), 182–207.

46 E. K. Chowdhury

Goetz, J. N., Brenning, A., Petschko, H., & Leopold, P. (2015). Evaluating machine learning and

statistical prediction techniques for landslide susceptibility modeling. Computers and

Geosciences, 81,1–11. https://doi.org/10.1016/j.cageo.2015.04.007

Hajek, P., & Abedin, M. Z. (2020). A proﬁt function-maximizing inventory backorder prediction

system using big data analytics. IEEE Access, 8, 58982–58994.

Hidayat, A. T., Dewantara, A. M. D., & Saifullah, S. (2020). The development of website on

management information system for e-commerce and services. Jurnal Sisfokom (Sistem

Informasi dan Komputer), 9(3), 380–386.

Horkoff, J., Borgida, A., Mylopoulos, J., Barone, D., Jiang, L., Yu, E., & Amyot, D. (2012,

September). Making data meaningful: The business intelligence model and its formal semantics

in description logics. In OTM Confederated International Conferences. On the move to mean-

ingful Internet systems (pp. 700–717). Springer.

Hutahayan, B. (2020). The mediating role of human capital and management accounting informa-

tion system in the relationship between innovation strategy and internal process performance

and the impact on corporate ﬁnancial performance. Benchmarking: An International Journal,

27(4), 1289–1318.

Kara, M. E., Fırat, S. Ü. O., & Ghadge, A. (2020). A data mining-based framework for supply chain

risk management. Computers & Industrial Engineering, 139, 105570.

Lau, H. C., Ip, A., Lee, C. K. M., & Ho, G. T. (2018). Development of a three-tier assessment

model: A case study. Benchmarking: An International Journal, 25(7), 2216–2229.

Munim, Z. H., Dushenko, M., Jimenez, V. J., Shakil, M. H., & Imset, M. (2020). Big data and

artiﬁcial intelligence in the maritime industry: A bibliometric review and future research

directions. Maritime Policy & Management, 47(5), 577–597.

Nguyen, T. T., Mia, L., Winata, L., & Chong, V. K. (2017). Effect of transformational-leadership

style and management control system on managerial performance. Journal of Business

Research, 70, 202–213.

Rodriguez-Galiano, V., Sanchez-Castillo, M., Chica-Olmo, M., & Chica-Rivas,

M. J. O. G. R. (2015). Machine learning predictive models for mineral prospectivity: An

evaluation of neural networks, random forest, regression trees and support vector machines.

Ore Geology Reviews, 71, 804–818.

Sabtu, A., Azmi, N. F. M., Sjarif, N. N. A., Ismail, S. A., Yusop, O. M., Sarkan, H., & Chuprat,

S. (2017, July). The challenges of extract, transform and loading (ETL) system implementation

for near real-time environment. In 2017 International Conference on Research and Innovation

in Information Systems (ICRIIS) (pp. 1–5). IEEE.

Shajalal, M., Hajek, P., & Abedin, M. Z. (2021). Product backorder prediction using deep neural

network on imbalanced data. International Journal of Production Research, 2021,1–18.

Ward, L., Agrawal, A., & Choudhary, A. (2016). A general-purpose machine learning framework

for predicting properties of inorganic materials. NPJ Computational Materials, 2, 16028. https://

doi.org/10.1038/npjcompumats.2016.28

Xin, Y., Kong, L., Liu, Z., Chen, Y., Li, Y., Zhu, H., Gao, M., Hou, H., & Wang, C. (2018).

Machine learning and deep learning methods for cybersecurity. IEEE Access, 6, 35365–35381.

Zhang, X. (2021). Application of data mining and machine learning in management accounting

information system. Journal of Applied Science and Engineering, 24(5), 813–820.

The Impact of Big Data on Accounting

Practices: Empirical Evidence from Africa

Mandella Osei-Assibey Bonsu, Naheed Roni, and Yongsheng Guo

Abstract Big data is much more than accounting and ﬁnancial data. Big data

including ﬁnancial and non-accounting data have become accessible in immense

volumes in distinct forms and in real time. The use of big data for accounting is

immobile in initial periods. However, academics have predicted that having high-

quality accessible and accelerated in real time might lead to more comprehensive

ﬁnancial reporting. Literature on big data is inconclusive, theoretical, and dearth

empirical studies and models. This prompted us to explore the impacts of big data on

accounting using accountants in an African emerging country, Nigeria. We use

multiple regression for 151 responses. The samples were collected using a random

sampling method. The results of the evidence show that big data has a positive and

signiﬁcant impact on ﬁnancial reporting, performance management, corporate

budgeting, audit evidence, risk management, and fraud management. Moreover,

evidence indicates that while big data signiﬁcantly impact accounting and auditing

of accountants, utilizing the diversity of data volume, data variety, and data velocity

signiﬁcantly enhances it. The study can help accountants, prospective accountants,

and accounting graduates hone their competencies in studying and producing big

data analytics, which will beneﬁt the industry. Moreover, business institutions of

higher learning should create business curriculums that use big data in their offer-

ings. Finally, policymakers can help by establishing governance models for big data

to organize its usage and prevent its exploitation.

Keywords Big data · Accounting · Auditing · Financial reporting · Nigeria · Africa

M. O.-A. Bonsu · N. Roni (✉) · Y. Guo

Department of Finance, Performance and Marketing, Teesside University International Business

School, Teesside University, Middlesbrough, Tees Valley, UK

e-mail: m.osie-assibeybonsu@tees.ac.uk;n.roni@tees.ac.uk;y.guo@tees.ac.uk

©The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

M. Z. Abedin, P. Hajek (eds.), Novel Financial Applications of Machine Learning

and Deep Learning, International Series in Operations Research & Management

Science 336, https://doi.org/10.1007/978-3-031-18552-6_4

48 M. O.-A. Bonsu et al.

1 Introduction

The dynamic business environment is calling business entities to invest time, money,

and efforts to adapt to envisage ways of doing things. In fact, the change of the entire

business model affects the way of accounting. Technology makes accounting free

from manual intervention and identiﬁes patterns and generates the exception reports,

leaving accountants with grey areas. As an emerged technical term, data is regarded

as the vehicle of the accounting profession (ICAEW, 2014). The growth of account-

ing and auditing has been empowering the development of big data to advance

technologies breakthroughs in multiple areas such as data analytics and Artiﬁcial

intelligence (Gepp et al., 2018; Bullock et al., 2020).

Big data is abundant more than accounting and ﬁnancial data (Petr & Abedin,

2020). Big data such as ﬁnancial and non-ﬁnancial data, accounting, and

non-accounting data, all of which become available in abundance volumes in distinct

forms and in actual time (Blazquez & Domenech, 2018; Bag et al., 2020; Basukie

et al., 2020). In fact, big data could enhance ﬁnancial accounting, reporting, and

auditing practices (Warren & Marz, 2015; Iqbal et al., 2020). This indicates that

having high-quality data available and processed in real time could lead to more

comprehensive ﬁscal information (Elmagrhi et al., 2019), improved management,

and more dependable budgeting. Furthermore, big data is thought to increase quality

by increasing accuracy and making information available in actual time (Cockcroft

& Russell, 2018).

In Nigeria, ﬁrms from every industry are at the frontline, experiencing ﬁrst-hand

the disruptive changes that affect their accountants. The rapid escalation in the

volume of data demands accountants to be equipped with the available technological

tools to analyze a much higher volume of data in their practice than has previously

been the case (Arner et al., 2015). For example, the use of data analytics hopes to

turn the accounting profession from a reactive and backward-looking exercise to a

constructive, continuous source of upward-looking insights that can be used all the

time, with the accountants as the custodian and translator of the underlying data

framework.

Insight on the impact of big data on accounting practices from accountants in

Nigeria are obtained for three reasons. First, ﬁntech in Nigeria has grown signiﬁ-

cantly for some years and is one of the ways for ﬁntech in Africa. Second, anecdotal

evidence that Nigeria is now home to over 200 ﬁntech ﬁrms, plus several ﬁntech

solutions offered by ﬁrms as part of the product portfolio. Nigeria’s sectors are

thriving and continue to amaze, exhibiting unwavering development and cutting-

edge data analytics. Finally, Nigeria has advanced as one of the African top ﬁntech

hubs, attracting 25 percent ($122 million) in investment raised by African tech

startups in 2019 (Disrupt Africa, 2021). In this paper, we examine the role of big

data in the practice of accounting and auditing in Nigeria.

Although some recent studies have linked growing technologies to the accounting

profession, there have been no scholarly empirical studies on the relationship

between big data and accounting (Chen et al., 2016; Shajalal et al., 2021). Although

some related literature studies have been conducted, there has been no empirical

research on the topic of accounting (Schmitz & Leoni, 2019; Lamboglia et al., 2020).

Furthermore, the application of big data for accounting is immobile at the early

stages (Scott & Orlikowski, 2012). Big data, however, is inconclusive, theoretical,

and dearth of empirical models. Therefore, more empirical studies are needed to

examine the impacts of big data on the works of accountants. To the best of our

knowledge, this is the ﬁrst study to examine the empirical impact of big data on

accounting and auditing practices evidenced from an Africa emerging economy,

Nigeria.

The Impact of Big Data on Accounting Practices: Empirical Evidence from Africa 49

The research provides contributions to the management accounting literature in

fourfolds. First, it is the ﬁrst empirical evidence to examine whether big data impacts

accounting practices in Nigeria. Second, the research contributes to the scant liter-

ature on big data and accounting practices in producing higher-quality audits to serve

existing purposes. Thus, this paper provides evidence of the signiﬁcance of big data

to auditing practices. Third, this research offers useful insights that may assist

accounting regulators in recognizing the importance of big data and accounting

relationships in developing accounting standards, as big data is seen as having the

ability to create and reﬁne accounting and auditing standards (Warren & Marz,

2015). Furthermore, the research could assist institutions of higher learning in

updating accounting curricula to handle big data. Finally, the study provides out-

comes that are more general with wider applicability by using an Africa emerging

country sample, which to the best of our knowledge, no research has studied.

The next section reviews the literature, followed by hypothesis development.

Section 3presents data and methods. Section 4reports the ﬁndings, followed by

discussions. The ﬁnal sections conclude with policy implications.

2 Literature Review

In recent times, big data has become the buzzword. Big data is described as high-

volume, high-velocity, and high-variety information assets that necessitate cost-

effective, novel data management to enable improved intelligence, decision-making,

and process automation (Gärtner & Hiebl, 2017). The three characteristics: volume,

velocity, and variety advocate that large volumes of transactions are created swiftly

from a diversity of sources.

Data are considered a vehicle for the accounting profession (ICAEW, 2014). On

the other hand, big data is abundant more than accounting and ﬁnancial data.

Financial and non-ﬁnancial data, accounting and non-accounting data, and numer-

ical and quantiﬁable data are all examples of big data, which is easily obtainable in

various formats, and in real time (Bag et al., 2020; Basukie et al., 2020). Big data has

the potential to enhance management accounting, ﬁnancial reporting, and ﬁnancial

accounting and auditing procedures (Brown-Liburd et al., 2015;Warren & Marz,

2015; Yoon et al., 2015; Iqbal et al., 2020). The study aims to investigate the impact

of big data on accounting and auditing, including big data on ﬁnancial reporting,

management performance, audit evidence, risk and fraud management, and corpo-

rate budgeting. Warren and Marz (2015) and Mofﬁtt and Vasarhelyi (2013)

suggested that big can enhance ﬁnancial reporting, improve transparency, account-

ing information quality, and enrich ﬁnancial reporting evidence.

50 M. O.-A. Bonsu et al.

However, the empirical evidence on the effectiveness of big data in accounting is

dearth in the literature. Apart from Al-Htaybat and von Alberti-Alhtaybat (2017),

Chen et al. (2015a,b), and Sardi et al. (2020), there is no empirical research on how

big data impact accounting, and auditing in Nigeria. Using interviews with 25 par-

ticipants, Al-Htaybat and von Alberti-Alhtaybat (2017), discovered that data ana-

lysts and accountants should work in conjunction to advance ﬁnancial reporting

utilizing data management. Sardi et al. (2020), on the other hand, found that

integrated performance grounded on big data can aid attain competitive advantage

for ﬁrms.

However, these studies were unable to determine whether there are empirical

positive relationships between big data, and accounting and auditing practices. The

approach informing in this study stresses the imperative of big data on accounting,

and auditing practices within accountants. Moreover, we have considered the

approach or credit risking (Abedin et al., 2018,2022) in relation to ﬁrms. Hence,

we expect empirical impacts of big data on each of accounting and auditing

practices. Researching the extant literature indicates that preceding studies about

big data and accounting are mainly theoretical, and there is a dearth of empirical

evidence on the use of big data in accounting. Moreover, no study has studied Africa.

3 Research Hypothesis

In this section, we develop hypotheses based on extant literature including big data

on accounting and auditing (ﬁnancial reporting, performance management, corpo-

rate budgeting, audit evidence, risk, and fraud management) as a results test if these

variables have positive relationships with big data.

3.1 Big Data and Financial Reporting Relationships

Transparency is the primary purpose of the governance system and corporate

reporting. Warren and Marz (2015) found that big data can increase transparency,

improve ﬁnancial reporting, and lead to improvements in accounting information

quality. Moreover, big data can enrich ﬁnancial reporting (Mofﬁtt & Vasarhelyi,

2013). The results of ﬁnancial accounting are ﬁnancial reporting that primarily

affects managers and stakeholders. However, corporate reporting does not address

the customers’changing needs.

Furthermore, in the era of big data, ﬁnancial reports are still made quarterly,

biannually, and annually. Financial reports are often publicly disclosed after the

audit at the end of the ﬁnancial year, which means that certain information may be no

longer relevant. Investors and stakeholders are increasingly awaiting fast ﬁnancial

data, perhaps daily. In this respect, one of the characteristics of big data is the speed

at which the data are processed and formed; big data schemes can now analyse and

produce data in actual period. This can facilitate companies’timely publication of

ﬁnancial reports. For example, Walmart, Amazon, and Royal Bank of Scotland have

used platforms for big data that process and provide data in real time (Marr, 2016).

As aresult, the implementation of a big-scale data system may have a signiﬁcant

impact on the ability of a company to provide timely ﬁnancial reports to the public.

The Impact of Big Data on Accounting Practices: Empirical Evidence from Africa 51

To date, there have been few empirical studies on big data and ﬁnancial reporting

relations. Aside from Al-Htaybat and von Alberti-Alhtaybat (2017), who found that

data analysts and accountants should collaborate to enhance ﬁnancial reporting

through advanced analytics (Yang et al., 2022). Moreover, Arnaboldi et al. (2017)

reviewed the literature and discovered from the literature that big data can help with

ﬁnancial reporting. Therefore, more empirical studies are needed to close this

signiﬁcant gap. Overall, big data can guarantee by escalating the quality of ﬁnancial

reporting, and henceforth hypothesized that:

H1: Big data is positively related to quality ﬁnancial reporting.

3.2 The Impact of Big Data on Performance Management

Through the collection, compilation, ﬁltering, analysis, interpretations, and dissem-

ination of appropriate data, performance is a set of measuring tools and dashboards

aimed at assessing management decisions and to quantifying the efﬁcacy and

efﬁciency of the actions conducted (Tambe, 2014). Many academics believe that

as competitiveness has increased, performance management has become increas-

ingly difﬁcult (Manyika et al., 2014). More organized and unstructured data are

becoming available and a diverse set of inputs is becoming increasingly vital for

long-term economic success. Information technology will provide different dimen-

sions to performance measurement processes. Typically, accounting managers use

structured data such as retention of employees, customer satisfaction surveys, and

return level to collect data on the four-point balanced scorecard (Richins et al.,

2017).

Accountants and ﬁnancial experts need to use large data to evaluate organiza-

tional performance (ACCA and IMA, 2013). First, Vera-Baquero et al. (2015)

present a big data resolution that can give ﬁrm analysts instantaneous acumens

into corporate performance and make measurements and signiﬁcant performance

indicators accessible. Second, an efﬁcient balance scorecard system requires exten-

sive and varied ﬁnancial and non-ﬁnancial data from internal and external sources.

Big data technologies can provide numerous and diverse customer data and allow

managers to effectively design BSC’s customers’perspectives, measures, objectives,

and strategies.

52 M. O.-A. Bonsu et al.

Studies on big data and performance management are mainly theoretical.

Elkmash et al. (2021) did a tentative investigation and discovered that big data

analytics lowers the cost of unstructured data analytics for customers and improves

the capacity to respond to consumer concerns quickly. Moreover, Sardi et al. (2020)

observed the relationships between big data and performance management and

found that big data might enhance competitive advantages. As a result, big data

can help managers establish the greatest vision and strategy for future occurrences.

The literature further determined that big data could help lengthen performance

measurement by creating novel performance indicators (Arnaboldi et al., 2017).

However, studies remain a theoretical argument in the absence of empirical research.

Therefore, we suggest that big data can positively enhance performance manage-

ment and accordingly propose the following hypothesis:

H2: Big data positively enhance the performance management of accountants.

3.3 Big Data and Corporate Budgeting Relationships

Budget is described as a quantitatively articulated realistic strategy for the future

(Gleim & Flesher, 2015). CIMA (2008) stated that a budget is a quantitative

description of a plan for a speciﬁc time. Budgets include anticipated returns and

sales, costs, reserve quantities, and expenditures, as well as liabilities, assets, and

ﬁnancial inﬂow (CIMA, 2008). However, budgeting is a management function

based on forecasts. According to Collier and Berry (2002), the budgeting process

often considers risk and uncertainty, as well as data on internal and external

occurrences. According to the Institute of Chartered Accountants of England and

Wales (ICAEW), accountants may use big data analytical models to enhance

budgeting and forecasting. Big data analytics is an organizational information

system that reduces uncertainty and better predicts future resource needs (Chen

et al., 2015a). However, Cokins (2014)claims that the use of advanced analytics

and big data in corporate operations has changed conventional costing planning and

budget variation control methods. Foremost, a large data volume provides managers

with many data inputs for budgeting, allowing them to create more accurate

budgeting valuations and predictions and hence lessen variances. Utilizing hundreds

of inputs instead of fewer can yield improved and further accurate projections in

forecasting (Duan & Xiong, 2015). Secondly, “Velocity,”will give data that are

analyzed simultaneously, allowing managers to track the budget implementation

process in real time, potentially reducing implementation errors. Data streaming,

conferring to Kudyba and Kudyba (2014) is one of the most important elements of

big data analytics. Real-time data streams from their source are analyzed and made

accessible to decision-makers. The third dimension, “Variety,”might offer a variety

of data formats for managers to choose from depending on the situation. Empirically,

analyzing the large quantity of data accessible on consumers’tastes, rivals’products,

and economic conditions with advanced analytics should produce more accurate

request and sales forecasts in actual time. This indicates that big data predictive

analytics could more properly estimate the future grounded on past events (Duan &

Xiong, 2015).

The Impact of Big Data on Accounting Practices: Empirical Evidence from Africa 53

Studies on the impact of big data on corporate budgeting are still based on theory

and dearth empirical evidence (Fisher et al., 2002; De Baerdemaeker & Bruggeman,

2015; Chen et al., 2016). Adding big data to the budgetary process can help manage

performance, resource allocation, and strategic target implementation with the least

amount of ﬂuctuation. Thus,

H3: Big data is positively improving corporate budgeting.

3.4 Big Data and Audit Evidence Relationships

The use of big data and analytics can help improve the efﬁciency and quality of

auditing (ICAEW, 2014). Audit evidence and big data relationships indicate con-

siderable convergence. Since it combines traditional evidence with reliable, sufﬁ-

cient, and relevant information (Yoon et al., 2015), increased transparency of audit

standards to audit evidence sources outside common ﬁnancial data. Hence, it is the

key facilitator for using big data by auditors. In fact, auditing conventional permit

auditors to gather evidence from any source and format if it beneﬁts in the formu-

lation of an opinion.

The International Standard on Auditing (500) coined audit evidence as any

information utilized by the auditor, whether presented in the accounting records or

vice versa. Moreover, AICPA (2004) reckoned that audit evidence is any informa-

tion utilized by the auditor to arrive at an audit conclusion, whether included in

accounting records or otherwise. This suggests that the ﬂexibility of auditing stan-

dards is in line with the distinctive features of big data. However, big data charac-

teristics can allow auditors to obtain evidence from a variety of sources, forms, and

in real time for the same audited items.

However, the motive is not only to have many diverse pieces of evidence, but also

for the evidence to be sufﬁcient, relevant, and reliable following auditing standards

(Alles, 2015; Brown-Liburd et al., 2015). The unique qualities of big data can

provide enough accurate audit evidence (Yoon et al., 2015). The accessibility of

large amounts of data in numerous formats and in real time, as well as the improved

competences of big data analytics, enhances the chances of collecting the most

adequate and relevant audit evidence. In summary, big data and related analysis

help auditors collect more appropriate relevant audit data and conclude an opinion

with a better level of assurance. However, to the best of our knowledge, no empirical

evidence is provided on whether big data positively improves the audit profession

via the big data audit evidence relations. Hence, the study hypothesized that:

H4: Big data is positively related to audit evidence.

54 M. O.-A. Bonsu et al.

3.5 Big Data and Risk and Fraud Management Relationships

Companies face a variety of risks that, if not properly assessed and handled, could

jeopardize their long-term viability. Among the main managerial concerns, and a key

governance necessity rule, is risk management. The board of directors of the ﬁrms

must maintain sound internal control and risk management systems (Council, 2011).

Bigdata can enhance risk surveillance, risk cover, and risk decision-making models

(Ibrahim et al., 2021). Big data and analytics offer accountants a variety of oppor-

tunities to improve risk management (ICAEW, 2014). Incorporating risk indicator

measurements will enhance the precision, and these indicators provide a predictive

value while providing the KRI in real time. However, because most risks are based

on the future, the more data available, the more precise the assessment and forecast

of risks. Big data predictive analytics enhances the stability and predictive perfor-

mance of risk assessment models, which allows managers to anticipate risk forecasts

more precisely (Duan & Xiong, 2015). Furthermore, big data can assist auditors to

measure the risks of their current or potential clients more precisely than ever,

including the risks of management fraud, falsiﬁcation of ﬁnancial statements, bank-

ruptcy, and risks related to the design and execution of internal controls (Cao et al.,

2015). Aboud and Robinson (2020) discovered that data analytics may be used to

detect or prevent fraud.

Equally, managers and investors can use advanced risk assessment and estimate

analytics to safeguard their companies and assets from ﬁnancial and market risks

such as liquidity, foreign currency, and share price volatility. Aside from fraud

detection, big data’s exceptional characteristics could aid enhance risk assessment,

measurement, and prediction. For instance, data volume and diversity will provide a

vast amount of internal, external, ﬁnancial, and non-ﬁnancial data in a range of

categories, resolving the data scarcity Chen et al. (2015b) studied the Alibaba Group

and found that big data can monitor and assess fraud threats in real time and send out

alerts to prevent fraud. Empowered with this, more studies are needed on how big

data may help with fraud detection and prevention (Cockcroft & Russell, 2018;

Aboud & Robinson, 2020). In fact, ﬁrms have begun to utilize big data resolutions to

develop their risk management schemes empirically.

However, there is a dearth of academic empirical studies on the use of big data in

enhancing risk management systems. Chen et al. (2015b) is the only empirical

research that we have found to bring the best out of our knowledge; hence, more

empirical evidence is needed to study the connections between big data, risk, and

fraud management. Hence, the study proposes that:

H5: Big data positively improves risk and fraud management.

The Impact of Big Data on Accounting Practices: Empirical Evidence from Africa 55

3.6 Research Framework

The research model that derives the analysis in this study is based on the empirical

review above. Therefore, the explanatory variable is big data, and the ﬁve hypoth-

eses discussed above form the basis of the empirical analysis of the research. The

research model for this study is presented in Fig. 1.

4 Research Methods and Data

4.1 Population and Sample

The paper examined the impact of big data on accounting practices among accoun-

tants in Nigeria. The sample consists of chartered accountants in Nigeria with an

international designation granted including ACCA and CIMA. We used the random

sampling technique which allowed us to obtain a sample of 152 representing

chartered accountants in Nigeria. The evidence-chartered accountants used as sam-

ple is empowered that it is vital for CAs to be sure of and have working knowledge of

big data. Furthermore, Nigeria, considered as the biggest economy in the African

continent, has grown in data science. Most ﬁrms have started to implement the tools

and techniques used in data science and ﬁntech. Hence, Nigeria presents a rich

setting to explore the empirical impacts of big data on accounting and auditing

practice.

H5 (+)

Volume

Variety

Velocity

Big Data

Financial

Reporting

Performance

Management

Corporate

Budgeting

Audit Evidence

Risk and Fraud

Management

H1 (+)

H2 (+)

H3 (+)

H4 (+)

Fig. 1 Study model

56 M. O.-A. Bonsu et al.

4.2 Questionnaires

Data were gathered among CAs through self-made questionnaires and were admin-

istered online from the period of October 2021 to January 2022. We prepared,

pre-tested, and revised the draft of the three-page, and two-section questionnaires.

First, pilot and pretesting were conducted by sending to 3 chartered accountants, and

2 University senior lecturers in accounting at UK recognized to the authors in big

data. They were requested to review, correct, and suggest improvements of the

original draft for relevance, content, and wordings. Second, we sent the reﬁned,

revised, and pre-tested questionnaires to respondents. The sections of the survey

asked CAs to comment on the impact of big data on ﬁnancial reporting, performance

management, risk and fraud management, corporate budgeting, and audit evidence,

and their respective proﬁles. To improve the response rate, cover letter was included

stating the survey objectives, deﬁning big data, and conﬁdentiality were guaranteed.

Finally, the survey link was generated online and sent in the email of selected

respondents, which assured that their responses would be completely anonymous.

4.3 Measurement of Big Data

For the measuring scales for the construct of big data, we relied on the existing

literature. Our study argues that the three big data characteristics (data volume, data

variety, and data velocity) are essential, since combined together contribute to the

big data constructions in accounting and auditing (Ghasemaghaei & Calic, 2019).

Hence, we asked 9 questions on big data regarding volume, variety, and velocity on

7 Likert scale from (1, strongly disagree to, 7, strongly agree).

4.4 Measurement of Accounting and Auditing Practices

We used ﬁnancial reporting, performance management, Risk & Fraud Management,

Corporate budgeting, and audit evidence as constructs to measure accounting and

auditing practices. Our self-administered questionnaires on accounting use twenty-

two (22) items on 7 Likert scale from (1, strongly disagree to, 7, strongly agree).

4.5 Methods

To examine the proposed hypotheses, we assessed the equations for the data. We

used regression as the current estimator for the impacts of big data on accounting.

The model is given as:

The Impact of Big Data on Accounting Practices: Empirical Evidence from Africa 57

Y=β0þβ1X1þβ2X2þβ3X3þεt,ð1Þ

where Yrepresents accounting and auditing practice practices, and X

1–3

represents

big data, respectively. In the ﬁrst place, we tested the effect of big data on Financial

Reporting (FR) by introducing volume, variety, and velocity. Hence,

FR =β0þβ1VLM1þβ2VRT2þβ3VCT2þεt:ð2Þ

Second, we tested the effect of big data on performance management, accord-

ingly, we regressed the model as:

PM =β0þβ1VLM1þβ2VRT2þβ3VCT2þεt:ð3Þ

We further tested the impact of big data volume, variety, and velocity on

Corporate Budgeting, we thus estimate as:

CB =β0þβ1VLM1þβ2VRT2þβ3VCT2þεt:ð4Þ

In addition, the single effect of big data on Audit Evidence was tested as follows:

AE =β0þβ1VLM1þβ2VRT2þβ3VCT2þεt:ð5Þ

Finally, we tested the effect of big data on Risk and Fraud management and

Eq. (5) shows as:

RFM =β0þβ1VLM1þβ2VRT2þβ3VCT2þεt:ð6Þ

For details description of variables, see Table 1. Following the distribution of the

questionnaires, we received completed one hundred and ﬁfty-two (165) out of three

hundred (300) distributed to a sample of accountants in Nigeria. After removing the

missing and incomplete data, we were left with 151 responses that were detailed and

adequate for analysis, accounting for 50.3 percent of the total. Table 2reports the

proﬁle of the respondents. We discovered 95 accountants, 62.91% of whom were

males and 37.09% of whom were females. Most of the respondents (54.30%) were

between the ages of 26–45, with 82.12% are qualiﬁed from the Institute of Chartered

Accountants of Nigeria (ICAN), followed by 10.59% with ACCA, and the majority

(41.04%) had worked between 6–10 years.

Besides, we found that 61.59% works for the banking, ﬁnance, and insurance,

23.18% for the service industry, and 15.23% for the manufacturing industry. Finally,

most of the respondents works in the private sector representing 75.50% leaving

24.50% for the public sector.

Common Method Bias

The study questionnaires are subjected to Common Method Bias (CMB) testing.

Because the study used a survey to acquire data from a single provider, there is still a

(continued)

58 M. O.-A. Bonsu et al.

Table 1 Description of variables

Constructs Variable Source

(7-point Likert scale from “strongly disagree”to

“strongly agree”)

Volume Larger amounts of data are analyzed. Ghasemaghaei and

Calic (2019)

The amount of data we examine is excessive.

We use a great deal of data, “in my opinion”

Velocity We are fast in exploring data Ghasemaghaei and

Calic (2019)

We analyze data quickly

We analyze different sources of data to gain insights

Variety We examine data from multitude of sources. Ghasemaghaei and

Calic (2019)

We use data to improve accounting information quality,

and ensures transparency

We use data to enrich reporting information, and perfor-

mance management.

(7-point Likert scale from “strongly disagree”to

“strongly agree”)

Financial

reporting

We use data to improve accounting information quality. Developed

We use data to enrich reporting information

We use data to ensure transparency, and improve

accounting information quality

We use data to improve performance management

Performance

management

We use structured data to assess organizational

performance

Developed

Big data may supply enormous and diverse customer

data

BDA allows to effectively design customer perspective objectives, measures,

targets, and strategies

BDA gives real-time insights and makes measurements

and key performance indicators

BDA provides business analytics real-time insights

Corporate

budgeting

Data analytics predicts models to improve budgeting and

forecasting

Developed

Data analytics provides managers with several inputs for budgeting, and allows

budget estimations

Managers could track budget implementation budget in

time

Developed

Audit evidence

Extend the scope of initiatives and compare them to

wider populations

Data may be analyzed in larger volumes and faster to

provide auditors with relevant insights.

BDA helps ﬁnancial auditors streamline the reporting

process

The Impact of Big Data on Accounting Practices: Empirical Evidence from Africa 59

Table 1 (continued)

Constructs Variable Source

Data analytics helps to detect fraud

(7-point Likert scale from “strongly disagree”to

“strongly agree”)

Risk and fraud

monitoring

Data can increase risk monitoring Developed

Data can enhance risk coverage, and creation of risk decisions making models

Analytics presents accountants with several chances to

improve risk management

Data analytics may be used to detect or prevent fraud

Big data analytics aid to improve risk assessment, pre-

diction, and measurement

Table 2 Summary of data from the respondents

Proﬁle Dimension Frequency n=151 Percentage (%)

Sex

Male 95 62.91

Female 56 37.09

Age

20–25 30 19.86

26–35 82 54.30

36 above 39 25.84

Education

Bachelors 54 35.76

Postgraduate 97 64.24

Certiﬁcation

ICAN 124 82.12

ACCA-UK 16 10.59

CIMA-UK 11 7.29

Experience

1–5 years 30 19.86

6–10 years 62 41.04

11-above years 59 39.1

Industry

Manufacturing 23 15.23

Banking, ﬁnance, insurance 93 61.59

Service 35 23.18

Sector

Public 37 24.50

Private 114 75.50

potential for CMB. As a result, the Harman single factor technique was applied,

which found 35 percent less than the 50 percent requirement. This suggests that the

constructs utilized in the study have no common method bias. According to the

ﬁndings, the data used in the study had no CMB concerns.

60 M. O.-A. Bonsu et al.

Measurement Models

To ensure model ﬁt and generate standardized loadings across constructs and items,

as well as between each of set of variables, we built a measurement model. Hence, it

is important to run a convergent and discriminant validity test prior to estimating

values using multiple regression to ensure the appropriateness of the measurement

model. From the results (Table 3), construct factor loading is higher than 0.7,

Cronbach alpha, and composite reliability (greater than the threshold 0.7) imply

strong reliability (Lance et al., 2006). Furthermore, the ﬁrst-order reﬂective items

composite reliability was robust and far above 0.8 (CR =0.944), Table 3), showing

high-scale dependability.

However, the values of the average variance estimates (AVE) were between 0.55

and 0.65, which were higher than the acceptability limit of 0.5. This indicates that the

variations recorded by the questionnaire items were substantially greater than the

changes caused by measurement error (Raykov, 2012). The convergent validity of

all three constructs was likewise supported, as seen in Table 3. As a result, the

underlying concept can account for more than half of the variance in the observed

variable (Hulland, 1999).

The correlations among each set of variables remained in the range between 0.27

and 0.45 (Table 4). Any highly correlated constructs higher than 0.90 could indicate

a common method bias (Bagozzi et al., 1991). All the relationships in our study are

less than 0.90. Therefore, we believe that multiple regression is adequate for the

study model.

We further employed the Fornell and Larker AVE metric to examine the dis-

criminant validity. The square root of the average variance estimates (AVE) of the

latent variable should be greater than the correlations across dimensions in the model

to meet the discriminant validity criteria. The square root of AVE for all constructs

(Table 5) is higher than their correlations (Table 4). Hence, discriminant validity was

found between the two conceptions. However, all AVE square roots were larger than

the correlations among all variables (evidence in Table 3). Hence, the study accepts

discriminant validity.

5 Empirical Results and Findings

Our study explored the impact of big data on accounting and auditing of accountants

in Nigeria. We used multiple regression estimates to test the hypotheses due to the

limited number of data sets (Eckstein et al., 2015). First, we examined the inﬂuence

Main variables Mean CR AVE

(continued)

The Impact of Big Data on Accounting Practices: Empirical Evidence from Africa 61

Table 3 Results of convergent and discriminant validity

Std.

Dev.

Factor

loadings

Cronbach

Alpha

Big data AVE (0.652) 6.295 0.305 0.826 0.944

Larger amounts of data, in my opinion,

are analysed

0.847

The amount of data we examine is

excessive.

0.799

We use a great deal of data, in my

opinion

0.802

We are fast in exploring data 0.802

We analyse data quickly 0.802

We analyse different sources of data to

gain insights

0.806

We examine data from multitude of

sources.

0.806

We use data to improve accounting

information quality and ensure trans-

parency

We use data to enrich reporting infor-

mation, and performance management

0.806

0.799

Accounting and auditing practice

AVE (0.659)

6.295 0.514 0.886 0.875

Financial reporting 6.217 0.433 0.743 0.865 0.616

We use data to improve accounting

information quality.

0.798

We use data to enrich reporting

information

0.732

We use data to ensure transparency,

and improve accounting information

quality

0.789

We use data to improve performance

management

0.818

Performance management 6.236 0.459 0.751 0.863 0.558

We use structured data to assess orga-

nizational performance

0.723

Big data may supply enormous and

diverse customer data

0.790

BDA allows to effectively design cus-

tomers perspective objectives, mea-

sures, targets, and strategies

0.794

BDA gives real-time insights and

makes measurements and key perfor-

mance indicators

0.714

BDA provides business analytics real-

time insights

0.711

Corporate budgeting 6.221 0.437 0.756 0.783 0.546

Data analytics predicts models to

improve budgeting and forecasting

0.756

Main variables Mean CR AVE

62 M. O.-A. Bonsu et al.

Table 3 (continued)

Std.

Dev.

Factor

loadings

Cronbach

Alpha

Data analytics provides managers with

several inputs for budgeting, and

allows budget estimations

0.749

Managers could track budget imple-

mentation budget in time

0.712

Audit evidence 6.247 0.500 0.790 0.866 0.564

Extend the scope of initiatives and

compare them to wider populations.

0.732

Data may be analyzed in larger vol-

umes and faster to provide auditors

with relevant insights.

0.730

BDA helps ﬁnancial auditors stream-

line the reporting process

0.752

Data analytics helps detect fraud 0.745

Overall, data analytics can aid to collect

more suitable and relevant evidence

0.793

Risk and fraud management 6.277 0.477 0.749 0.884 0.559

Data can increase risk monitoring 0.747

Data can enhance risk coverage, and

creation of risk decisions making

models

0.744

Analytics presents accountants with

several chances to improve risk man-

agement

Data analytics may be used to detect or

prevent fraud

Big data analytics aid to improve risk

assessment, prediction, and

measurement

0.749

0.748

0.750

0.751

Table 4 Correlation results

CA AVE Big FRep PMgt. CBugt. AEvid. RFMgt.

Big data 0.826 0.652

FRep. 0.743 0.616 0.338

PMgt. 0.751 0.558 0.347 0.450

CBugt. 0.756 0.554 0.351 0.342 0.285

AEvid. 0.790 0.564 0.356 0.352 0.352 0.325

RFMgt. 0.749 0.559 0.360 0.387 0.384 0.471 0.271

Variable

of each data volume, data variety, and data velocity on accounting and auditing

practice and explored their effects together.

The Impact of Big Data on Accounting Practices: Empirical Evidence from Africa 63

Table 5 Variables, Cronbach Alpha, and AVE square root

Variable Cronbach Alpha Average variance estimate Square root AVE

Big data 0.826 0.652 0.81

FRep. 0.743 0.616 0.78

PMgt. 0.751 0.558 0.751

CBugt. 0.756 0.554 0.744

AEvid. 0.790 0.564 0.751

RFMgt. 0.749 0.559 0.75

Table 6 Results of big data, accounting, and auditing relationships

FRep PMgt. CBugt. AEvid. RFMgt.

Model 1 Model 2 Model 3 Model 4 Model 5

Big data 0.345

(0.00)***

0.432 (0.00)*** 0.333

(0.00)***

0.378

(0.00)***

0.360

(0.00)***

Notes: The table presents the results of big data, accounting, and auditing relationships. Big data,

FRep, PMgt, CBudgt, AEvid, RFMgt represent big data, Financial Reporting, Performance and

Management, Control budgeting, Audit Evidence, Risk and Financial Management, ***, **, *

indicate signiﬁcance at 1%, 5%, and 10% level, the p-value is provided in the parathesis

5.1 Results of Big Data, Accounting, and Auditing

Relationships

Table 6provides estimates highlights and empirical ﬁndings on the impact of big

data on accounting and auditing using the multiple regression model employed. The

results indicate that big data is statistically positive and signiﬁcant in ﬁnancial

reporting (β=0.345, p-value =0.000). Hence, H1 is approved. Likewise, the use

of big data is positive and signiﬁcant in performance management (β=0.432, p-

value =0.000), conﬁrming H2. Moreover, big data is positive on corporate

budgeting (β=0.333, p-value =0.00), supporting H3, big data is positive and

signiﬁcant on audit evidence (β=0.378, p-value =0.000), risk and fraud manage-

ment (β=0.360, p-value =0.000). These ﬁndings support Hypothesis H4-H5 are

further discussed in the Conclusions and Discussions sections.

The R

of endogenous constructions value for ﬁnancial reporting, performance

management, corporate budgeting, audit evidence and risk and fraud management is

0.54, 0.55, 0.51, 0.52, and 0.52, respectively, which indicates that the model is

responsible for roughly 54%, 55%, 51%, 52%, and 52% of the volatility in account-

ing and auditing practices. The summary showing which of the hypotheses were

approved is also presented in Table 8.

Variable

64 M. O.-A. Bonsu et al.

Table 7 Results of data variety, variety, velocity, accounting, and auditing relationships

FRep PMgt. CBugt. AEvid. RFMgt.

Model 1 Model 2 Model 3 Model 4 Model 5

Volume 0.116

(0.00)***

0.149 (0.00)*** 0.103

(0.00)***

0.126

(0.00)***

0.120

(0.00)***

Variety 0.146

(0.00)***

0.142

(0.00)***

0.110

(0.00)***

0.126

(0.00)***

0.120

(0.00)***

Velocity 0.126

(0.00)***

0.141

(0.00)***

0.120

(0.00)***

0.126

(0.00)***

0.120

(0.00)***

Notes: The table presents the results of data characteristics. Big data, FRep, PMgt, CBudgt, AEvid,

RFMgt represent big data, Financial Reporting, Performance and Management, Control budgeting,

Audit Evidence, Risk and Financial Management, ***, **, * indicate signiﬁcance at 1%, 5%, and

10% level, the p-value is provided in the parathesis

Table 8 Hypothesis testing

Hypothesis Relationships Total estimates Percentage Prove

H1 Big→FRep 0.345 0.000*** Approved

H2 Big→PMgt. 0.432 0.000*** Approved

H3 Big→CBugt. 0.333 0.000*** Approved

H4 Big→AEvid. 0.378 0.000*** Approved

H5 Big→RFMgt. 0.360 0.000*** Approved

Notes: The table presents the hypothesis for the study. Big, FRep, PMgt, CBudgt, AEvid, RFMgt,

represent big data, Financial Reporting, Performance and Management, Control budgeting, Audit

Evidence, Risk and Financial Management, ***, **, * indicate signiﬁcance at 1%, 5%, and 10%

level, the p-value is provided in the parathesis

Table 7reports highlights of the estimations and empirical evidence from the

models employed. From the results, the volume is positive on ﬁnancial reporting at a

signiﬁcance level of 1%. Similarly, variety is positive and signiﬁcant at the 1% level

in ﬁnancial reporting, and velocity is positively related and signiﬁcant in ﬁnancial

reporting. The positive impact suggests that the use of big data will signiﬁcantly

improve the ﬁnancial reporting of accountants by about 0.345%. The results conﬁrm

with Marr (2016) suggestion that implementing big data system may strongly affect

ﬁrm capacity to timely disclose ﬁnancial reporting. With regard to performance

management, there is evidence of positive and signiﬁcant impact of volume, variety,

and velocity on performance management at signiﬁcance level of 1%. This suggests

that accountants use high levels of data volume, data velocity, and data variety have

the best means to assess ﬁrm performance. The results, however, conﬁrm with Sardi

et al. (2020) who found that big data might help organization attain competitive

advantage. Moreover, Elkmash et al. (2021) found that big data analytics lowers the

cost of unstructured data analysis for customers and improves the capacity to

respond to consumer concerns swiftly. As results, big data can help managers

establish the greatest vision and strategy for future occurrences.

From model 3, volume, variety and velocity are positive and signiﬁcant on

corporate budgeting. The positive impact on corporate budgeting indicates

high-level use of high levels of data velocity, data volume, and data variety leads

0.333 percent in corporate budgeting of accountants in Nigeria. The result afﬁrms

that accountants use the predictive model of large data to improve budget and

forecasting (ICAEW, 2020). For example, a large data volume provides accountants

and managers with many inputs for budgeting, allowing them to create more

accurate budgeting estimations and predictions and hence reduce variances. How-

ever, the result is novel and contributes to the extant literature as studies on big data

impact on corporate budgeting are still based on theory and dearth empirical

evidence (Fisher et al., 2002; De Baerdemaeker & Bruggeman, 2015; Chen et al.,

2016).

The Impact of Big Data on Accounting Practices: Empirical Evidence from Africa 65

With respect to model 4, the results discovered a positive and signiﬁcant effect of

data volume, data velocity and data variety on audit evidence at 1% signiﬁcance

level value suggesting that, accountants of Nigeria ﬁrms high level use of big data

enhance audit evidence at a coefﬁcient of 0.378%. The results conﬁrm with ICAEW

(2014) that the use of big data and analytics could help improve the quality and

efﬁciency of auditing. Between big data and audit evidence, there is a consideration

convergence, and therefore big data will play an essential role in auditing. Therefore,

unique qualities of big data can provide sufﬁcient and accurate audit evidence (Yoon

et al., 2015). However, no empirical evidence is provided, and there this ﬁnding

contributes to the knowledge base.

Finally, the results (model 5) evidence positive and signiﬁcant impact data

volume, velocity, and variety on risk and fraud management at 1% signiﬁcance

level value. The coefﬁcient magnitude indicates that high-level use of data volume,

data variety and data velocity will lead increased risk and fraud monitoring at 0.36%.

The result conﬁrms with Chen et al. (2015b), who found that Alibaba Group’s big

data system can monitor and assess fraud threats in real time and send out alerts to

prevent fraud. Moreover, the volume, variety, and velocity of high levels of data

could help improve risk assessment, prediction, and measurement. For example,

volume and variety will supply a large amount of internal, external, ﬁnancial, and

non-ﬁnancial data in various formats, overcoming the data shortage issue (Table 8).

6 Discussion on Results

The results of this study present that while big data signiﬁcantly impact accounting

and auditing of accountants, utilizing the diversity of data (i.e., data volume, data

variety, and data velocity) signiﬁcantly improves it. This indicates that analyzing

data from both multiple sources yields economically valuable insights, focusing on

swiftly processing data or analyzing large volumes, variety, and velocity does

necessarily provide ﬁnancial beneﬁts for accountants and auditors. The results

indicate that big data has signiﬁcant positive impact on ﬁnancial reporting. The

results conﬁrm Marr (2016), who suggested that the implementation of a big data

system has a major effect on ﬁrm capacity to provide timely ﬁnancial reporting to the

public. However, the ﬁnding supports Warren and Marz (2015) who found that big

data can enhance ﬁnancial reporting and enrich. Moreover, Mofﬁtt and Vasarhelyi

(2013) established that big data enrich ﬁnancial reporting information. The ﬁnding

suggests that accountants could improve the quality and accuracy of ﬁnancial

reports, especially when big data and continuous analytics is used.

66 M. O.-A. Bonsu et al.

The ﬁndings further ﬁnd a signiﬁcant positive impact of big data on performance

management. The ﬁnding is consistent with Sardi et al. (2020), who indicated that

big data improve competitive advantage. Besides, ACCA and IMA (2013) asserted

that big data used by accountants and ﬁnance experts is paramount to examine ﬁrm

performance. Hence, big data can help managers establish the greatest vision and

strategy for future occurrences. Moreover, the results indicate a signiﬁcant positive

impact of big data on corporate budgeting. The results afﬁrm that the more data

obtainable and more reliable an organization revenue and expenses, the more

effective a static budget is at delivering useful information for decision-making

and predict future budgets. However, the result is novel and contributes to the

literature, as studies of the relationship between big data and corporate budgeting

is still theoretical (De Baerdemaeker & Bruggeman, 2015; Chen et al., 2016).

However, the results found a signiﬁcant and positive impact on the audit evidence.

The results support the notion that the accessibility of large amounts of data in

various formats and simultaneously, as well as the improved competences of big data

analytics, enhances the chances of collecting the most adequate and relevant audit

evidence. Finally, the results show that big data has a signiﬁcant and positive effect

on risk and fraud management. The ﬁnding is in line with Chen et al. (2015b) who

found that Alibaba Group in China big data system can monitor and assess fraud

threats in real time and send out alerts to prevent fraud. This suggests that big data

can increase risk coverage, risk monitoring, and creation of risk decision-making

models, permeating managers to anticipate risk forecasts more precisely (Duan &

Xiong, 2015).

To further explore the effect of each big data dimensions on accounting and

auditing practices, our study assessed data volume, variety, and velocity when

accountants utilized diverse levels of big data dimensions. The ﬁndings show that

although accountants use high levels of data volume, velocity, and variety regarding

their accounting and auditing practices, data variety has the highest means regarding

accounting and auditing practices. The result is scholarship (theoretically and prac-

tically) signiﬁcant, with the assumption that one needs to have a farther comprehen-

sion of effect of big data on accountants.

6.1 Theoretical Contribution

Academics and the literature view big data as a vehicle for the accounting profession

(ICAEW, 2014) and have the potential to add value to companies and enhance their

performance. However, studies argue that big data is far more than accounting data.

Moreover, big data have potential to advance management accounting, ﬁnancial

reporting, ﬁnancial accounting procedures, and auditing (Iqbal et al., 2020).

Researching the extant literature indicates that preceding studies on big data and

accounting are mainly theoretical. Therefore, the empirical study on the effect of big

data in accounting is dearth in the literature. However, to the best of our knowledge,

there are no empirical studies that investigated the impact of big data on accounting

and auditing practice in emerging markets. Moreover, no study has studied Africa.

As such, the role of big data utilizing in enhancing accounting and auditing works is

not well understood. The gap is what our study examined. To address the study

objectives, we surveyed chattered accountants from the African emerging economy,

Nigeria to examine the impact of big data on accounting and auditing practice. We

make numerous theoretical contributions.

The Impact of Big Data on Accounting Practices: Empirical Evidence from Africa 67

1. We underline the need for accountants and managers adopting big data to publish

high-quality information to lessen agency costs and vagueness from an agency

theory approach. We illustrated the need to theoretically distinguish between big

data dimensions when assessing their effects on accounting and auditing

methods, but it might be treated holistically.

2. The results extend to the understanding of the big data literature of the impact of

data volume, variety, and velocity on accounting and auditing. Generally, the

ﬁndings show that each big data dimension might have a different impact on

accounting and auditing procedures. Even though data volume, variety, and

velocity all have an impact on accounting and auditing, data variety has the

most impact. Our ﬁndings contribute to the big data literature by examining

how each of the big data’s three primary characteristics affects accountants and

auditors’work.

3. Big data has a large and positive impact on budgeting, as per the results.

However, because studies of the interaction between big data and corporate

budgeting are still theoretical (De Baerdemaeker & Bruggeman, 2015; Chen

et al., 2016), the result is novel and contributes to the literature.

4. A novel contribution of our work to the big data literature is the difference in the

inﬂuence of data volume, variety, and velocity on accounting and auditing. Our

ﬁndings represent the ﬁrst step in determining the effects of big data character-

istics on accounting and auditing in Africa’s emerging economy.

Furthermore, our study provides a signiﬁcant theoretical contribution by devel-

oping a measurement scale in the context of accounting and auditing. To summarize,

this is the ﬁrst empirical research to examine the effect of big data on accounting and

auditing in the African emerging economy. Moreover, this is also the ﬁrst to

empirically examine the relationships in Africa context.

6.2 Policy Implications

The study preceding discussions supplies the following implications. First, big data

can help develop accounting and overcome the constraints of numerous accounting

procedures in relation to the data. As a result, accountants, prospective accountants,

and accounting graduates should hone their competencies in studying and producing

big data analytics, which will beneﬁt the industry. Second, the study is important to

managers, since it shows how big data represents a hopeful future. Furthermore,

accounting teaching bodies have a strong demand for data analysis and data science

employment, and there is a lack of such jobs on the job market (Ibrahim et al., 2021).

As a result, business institutions of higher learning should create business curricu-

lums that use big data in their offerings. As an outcome of our results, prospective

accountants should have a thorough understanding of numerous business matters, as

well as a solid understanding of various big data features and how to apply them in

accounting and auditing operations. Finally, policymakers can help by establishing

governance frameworks for big data to organize its usage and prevent its

exploitation.

68 M. O.-A. Bonsu et al.

7 Conclusion, Limitations, and Further Studies

The main objective of our study was to close an indispensable gap in the literature

concerning the effectiveness of big data on accounting and auditing practice. The

study sampled respondents from Nigeria, which is an African emerging economy.

Results indicate that big data signiﬁcantly and positively improves ﬁnancial

reporting, performance management, audit evidence, corporate budgeting, risk,

and fraud management of accountants. Moreover, the study found that big data

positively and signiﬁcantly impact risk and fraud management. Interestingly, the

effect of data volume, data variety, and data velocity enhances accounting and

auditing practices. One of the unique contributions of this study is creating fascinat-

ing insights about the empirical impact of big data on accounting when accountants

use different characteristics of big data.

Albeit data volume, variety, and velocity could be signiﬁcant and positively

impact accounting and auditing, data variety has the strongest impact. Our results

add to the big data literature by investigating how each of the three main dimensions

of big data impacts the work of accountants and auditors. These ﬁndings assist

accountants in using big data analytics to help businesses obtain deeper insight,

anticipate future outcomes, and streamline non-routine processes. Furthermore, big

data presents prospects for the accounting profession to add value and assist busi-

nesses in transforming decision-making in a variety of ways.

There are some potential caveats to this study. First, this study employed a cross-

sectional survey to test statistical relations in the proposed study framework. We are

calling further studies to employ longitudinal approach as cross-sectional data are

inadequate to test the causal relations amid constructs in the study model. Second,

we selected participants through the random sample technique. Despite it was

considered necessary due to the nature of data received from the Nigerian market,

it has caveats in terms of generalizability of the conclusions. Finally, we call for

further studies to further validate the results of this study, as our study recruited

respondents from Nigeria. Empirical studies from advanced countries would be

helpful.

The Impact of Big Data on Accounting Practices: Empirical Evidence from Africa 69

References

Abedin, M. Z., Guotai, C., Colombage, S., & Moula, F. E. (2018). Credit default prediction by

using a support vector machine and a probabilistic neural network. Journal of Credit Risk, 14(2),

1–27. https://doi.org/10.21314/JCR.2017.233

Abedin, M. Z., Guotai, C., Hajek, P., Tong, Z., & Moula, F. E. (2022). Combining weighted

SMOTE with ensemble learning for the class-imbalanced prediction of small business credit

risk. Complex & Intelligent Systems. https://doi.org/10.1007/s40747-021-00614-4

Aboud, A., & Robinson, B. (2020). Fraudulent ﬁnancial reporting and data analytics: An explan-

atory study from Ireland. Accounting Research Journal, 35(1), 21–36.

Disrupt Africa. (2021). Disrupt Africa. Accessed from https://disrupt-africa.com/funding-report/

Al-Htaybat, K., & von Alberti-Alhtaybat, L. (2017). Big data and corporate reporting: Impacts and

paradoxes. Accounting, Auditing & Accountability Journal, 30(4), 850–873.

Alles, M. G. (2015). Drivers of the use and facilitators and obstacles of the evolution of big data by

the audit profession. Accounting Horizons, 29(2), 439–449.

American Institute of Certiﬁed Public Accountants (AICPA). (2004). Audit evidence. Statement on

auditing standards no. 106. AICPA.

Arnaboldi, M., Busco, C., & Cuganesan, S. (2017). Accounting, accountability, social media and

big data: Revolution or hype? Accounting, Auditing & Accountability Journal, 30(4), 762–776.

Arner, D. W., Barberis, J., & Buckley, R. P. (2015). The evolution of Fintech: A new post-crisis

paradigm. Georgetown Journal of International Law, 47, 1271.

Association of Chartered Certiﬁed Accountants (ACCA) and Institute of Management Accountants

(IMA). (2013). Big data: Its powers and perils. ACCA. Accessed from http://www.accaglobal.

com/content/dam/acca/global/PDF-technical/futures/pol-afa-bdpap.pdf

Bag, S., Wood, L. C., Xu, L., Dhamija, P., & Kayikci, Y. (2020). Big data analytics as an

operational excellence approach to enhance sustainable supply chain performance. Resources,

Conservation and Recycling, 153, 104559.

Bagozzi, R. P., Yi, Y., & Phillips, L. W. (1991). Assessing construct validity in organizational

research. Administrative Science Quarterly, 36(3), 421–458.

Basukie, J., Wang, Y., & Li, S. (2020). Big data governance and algorithmic management in

sharing economy platforms: A case of ridesharing in emerging markets. Technological Fore-

casting and Social Change, 161, 120310.

Blazquez, D., & Domenech, J. (2018). Big data sources and methods for social and economic

analyses. Technological Forecasting and Social Change, 130,99–113.

Brown-Liburd, H., Issa, H., & Lombardi, D. (2015). Behavioral implications of big data’s impact on

audit judgment and decision making and future research directions. Accounting Horizons, 29(2),

451–468.

Bullock, J., Young, M. M., & Wang, Y. F. (2020). Artiﬁcial intelligence, bureaucratic form, and

discretion in public service. Information Polity, 25(4), 491–506.

Cao, M., Chychyla, R., & Stewart, T. (2015). Big data analytics in ﬁnancial statement audits.

Accounting Horizons, 29(2), 423–429.

Chen, D. Q., Preston, D. S., & Swink, M. (2015a). How the use of big data analytics affects value

creation in supply chain management. Journal of Management Information Systems, 32(4),

4–39.

Chen, J., Tao, Y., Wang, H., & Chen, T. (2015b). Big data based fraud risk management at Alibaba.

The Journal of Finance and Data Science, 1(1), 1–10.

Chen, Y., Chen, H., Gorkhali, A., Lu, Y., Ma, Y., & Li, L. (2016). Big data analytics and big data

science: A survey. Journal of Management Analytics, 3(1), 1–42.

CIMA (Chartered Institute of Management Accounting). (2008). Budgeting: Topic gateway series

no. 27. CIMA. Available at: http://www.cimaglobal.com/Documents/ImportedDocuments/cig_

tg_budgeting_mar08.pdf

Cockcroft, S., & Russell, M. (2018). Big data opportunities for accounting and ﬁnance practice and

research. Australian Accounting Review, 28(3), 323–333.

70 M. O.-A. Bonsu et al.

Cokins, G. (2014). Mining the past to see the future: CFOs and their teams can use business

analytics to make better decisions. Strategic Finance, 96(5), 23–31.

Collier, P. M., & Berry, A. J. (2002). Risk in the process of budgeting. Management Accounting

Research, 13(3), 273–297.

Council, F. R. (2011). The UK corporate governance code. Financial Reporting Council.

De Baerdemaeker, J., & Bruggeman, W. (2015). The impact of participation in strategic planning

on managers’creation of budgetary slack: The mediating role of autonomous motivation and

affective organisational commitment. Management Accounting Research, 29,1–12.

Duan, L., & Xiong, Y. (2015). Big data analytics and business analytics. Journal of Management

Analytics, 2(1), 1–21.

Eckstein, D., Goellner, M., Blome, C., & Henke, M. (2015). The performance impact of supply

chain agility and supply chain adaptability: The moderating effect of product complexity.

International Journal of Production Research, 53(10), 3028–3046.

Elmagrhi, M. H., Ntim, C. G., Elamer, A. A., & Zhang, Q. (2019). A study of environmental

policies and regulations, governance structures, and environmental performance: The role of

female directors. Business Strategy and the Environment, 28(1), 206–220.

Elkmash, M. R. M., Abdel-Kader, M. G., & El Din, B. B. (2021). An experimental investigation of

the impact of using big data analytics on customers’performance measurement. Accounting

Research Journal.

Fisher, J. G., et al. (2002). Using budgets for performance evaluation: Effects of resource allocation

and horizontal information asymmetry on budget proposals, budget slack, and performance. The

Accounting Review, 77(4), 847–865.

Gärtner, B., & Hiebl, M. R. (2017). Issues with big data (pp. 161–172). Routledge.

Gepp, A., Linnenluecke, M. K., O’Neill, T. J., & Smith, T. (2018). Big data techniques in auditing

research and practice: Current trends and future opportunities. Journal of Accounting Literature,

40, 102–115.

Ghasemaghaei, M., & Calic, G. (2019). Does big data enhance ﬁrm innovation competency? The

mediating role of data-driven insights. Journal of Business Research, 104,69–84.

Gleim, & Flesher, D. L. (2015). Gleim CMA Review, Part. 1: Financial reporting, planning,

performance, and control. GLEIM Publications, Inc.

Hulland, J. (1999). Use of partial least squares (PLS) in strategic management research: A review of

four recent studies. Strategic Management Journal, 20(2), 195–204.

Ibrahim, A. E. A., Elamer, A. A., & Ezat, A. N. (2021). The convergence of big data and

accounting: Innovative research opportunities. Technological Forecasting and Social Change,

173, 121171.

ICAEW (Institute of Chartered Accountants in England and Wales). (2020). Big data and analytics

–what’s new? Chartered Accountants’Hall.

Institute of Chartered Accountants in England and Wales (ICAEW). (2014). Big data and analytics

- What’s new? Chartered Accountants’Hall.

Iqbal, R., Doctor, F., More, B., Mahmud, S., & Yousuf, U. (2020). Big data analytics: Computa-

tional intelligence techniques and application areas. Technological Forecasting and Social

Change, 153, 119253.

Kudyba, S., & Kudyba, S. (2014). Big data, mining, and analytics. Auerbach Publications.

Lamboglia, R., Lavorato, D., Scornavacca, E., & Za, S. (2020). Exploring the relationship between

audit and technology. A bibliometric analysis. Meditari Accountancy Research, 29(5),

1233–1260.

Lance, C. E., Butts, M. M., & Michels, L. C. (2006). The sources of four commonly reported cutoff

criteria: What did they really say? Organizational Research Methods, 9(2), 202–220.

Manyika, J., Chui, M., Brown, B., Bughin, J., Dobbs, R., Roxburgh, C., & Byers, A. H. (2014,

May). Big data: The next frontier for innovation, competition, and productivity. McKinsey

Global Institute.

Marr, B. (2016). Big data in practice: How 45 successful companies used big data analytics to

deliver extraordinary results. Wiley.

The Impact of Big Data on Accounting Practices: Empirical Evidence from Africa 71

Mofﬁtt, K. C., & Vasarhelyi, M. A. (2013). AIS in an age of big data. Journal of Information

Systems, 27(2), 1–19.

Petr, H., & Abedin, M. Z. (2020). A proﬁt function maximizing inventory backorder prediction

system using big data analytics. IEEE Access, 8, 58982–58994.

Raykov, T. (2012). Scale construction and development using structural equation modeling.

Richins, G., Stapleton, A., Stratopoulos, T. C., & Wong, C. (2017). Big data analytics: Opportunity

or threat for the accounting profession? Journal of Information Systems, 31(3), 63–79.

Sardi, A., Sorano, E., Cantino, V., & Garengo, P. (2020). Big data and performance measurement

research: Trends, evolution and future opportunities. Measuring Business Excellence.https://

doi.org/10.1108/MBE-06-2019-0053

Schmitz, J., & Leoni, G. (2019). Accounting and auditing at the time of blockchain technology: A

research agenda. Australian Accounting Review, 29(2), 331–342.

Scott, S. V., & Orlikowski, W. J. (2012). Reconﬁguring relations of accountability: Materialization

of social media in the travel sector. Accounting, Organizations and Society, 37(1), 26–40.

Shajalal, M., Petr, H., & Abedin, M. Z. (2021). Product backorder prediction with deep neural

network on imbalance data. International Journal of Production Research, 2021,1.

Tambe, P. (2014). Big data investment, skills, and ﬁrm value. Management Science, 60(6),

1452–1469.

Vera-Baquero, A., Palacios, R. C., Stantchev, V., & Molloy, O. (2015). Leveraging big-data for

business process analytics. The Learning Organization, 22(4), 215–228.

Warren, J., & Marz, N. (2015). Big data: Principles and best practices of scalable realtime data

systems. Simon & Schuster.

Yang, F., Qiao, Y., Abedin, M. Z., & Huang, C. (2022). Privacy-preserved credit data sharing

integrating blockchain and federated learning for industrial 4.0. IEEE Transactions on Indus-

trial Informatics.https://doi.org/10.1109/TII.2022.3151917

Yoon, K., Hoogduin, L., & Zhang, L. (2015). Big data as complementary audit evidence. Account-

ing Horizons, 29(2), 431–438.

Part II

Financial Risk Prediction Using Machine

Learning

Using Outlier Modiﬁcation Rule

for Improvement of the Performance

of Classiﬁcation Algorithms in the Case

of Financial Data

Md. Rabiul Auwul, Md. Ajijul Hakim, Fahmida Tasnim Dhonno,

Nusrat Afrin Shilpa, Ashrafuzzaman Sohag,

and Mohammad Zoynul Abedin

Abstract This study aims to improve the performance of Data Analytics

(DA) algorithms by mining outliers from credit card fraud detection datasets. In

doing so, we analyze the performance of data analytics algorithms, such as Linear

Discriminant Analysis (LDA), k-Nearest Neighbor (k-NN), Naïve Bayes (NB) and

Support Vector Machine (SVM), by comparing the original and modiﬁed datasets in

the absence and presence of outliers. To generate modiﬁed dataset, this chapter

proposes an outlier mining method based on Median (MED) and Median Absolute

Deviation (MAD). Performance measures such as accuracy, sensitivity, speciﬁcity,

detection rate, misclassiﬁcation error rate, AUC, and pAUC evaluate the perfor-

mance of the DA algorithms. Empirical ﬁndings show that the performance of the

DA algorithms on modiﬁed dataset shows better results than the original data for

both simulated dataset and real-life credit card datasets. This study offers new

insights into ﬁnancial decision makers and stakeholders in the credit card industry.

Keywords Financial data · Classiﬁcation · Outlier detection · Modiﬁcation

M. R. Auwul

Department of Mathematics, Faculty of Science and Technology, American International

University-Bangladesh, Dhaka, Bangladesh

M. A. Hakim

Foreign Exchange and Remittance Department, Travelex Qatar, Golbex Business Center, Doha,

Qatar

F. T. Dhonno · N. A. Shilpa · A. Sohag

Department of Finance and Banking, Hajee Mohammad Danesh Science and Technology

University, Dinajpur, Bangladesh

M. Z. Abedin (✉)

Department of Finance, Performance and Marketing, Teesside University International Business

School, Teesside University, Middlesbrough, Tees Valley, UK

e-mail: m.abedin@tees.ac.uk

©The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

M. Z. Abedin, P. Hajek (eds.), Novel Financial Applications of Machine Learning

and Deep Learning, International Series in Operations Research & Management

Science 336, https://doi.org/10.1007/978-3-031-18552-6_5

76 M. R. Auwul et al.

1 Introduction

The competition condition of credit markets has altered since ﬁnancial technology

companies (FinTechs) and giant technology companies (BigTechs) have initiated

providing alternative credit (Kowalewski & Pisany, 2022). Since the last two

decades, ﬁnancial institutions have undergone extensive ﬁnancial technologies

which have brought changes in credit provision, information, savings, communica-

tion, transactions, and cyber security (Murinde et al., 2022). Machine learning,

predictive analytics, data science, and artiﬁcial intelligence are all widely used by

FinTech organizations to facilitate ﬁnancial decision-making, as well as eliminate

credit default risks (Abedin et al., 2022).

Data science (DS) plays a vital role in managing credit default risk and detecting

fraud in ﬁnancial decisions. DS refers to the process of categorizing a set of tested

elements, such as bonds, enterprises, stocks, countries, and so on, into predetermined

similar groups (Kulczycki & Franus, 2020). DS algorithms predict credit and fraud

risks quickly, helping to increase the efﬁciency of ﬁnancial decision-making pro-

cesses (Hajek & Abedin, 2020). To improve ﬁnancial decision-making performance,

there is a wide range of DS algorithms such as Naïve Bayes (NB), k-Nearest

Neighbor (k-NN), Linear Discriminant Analysis (LDA), and Support Vector

Machine (SVM) (Li et al., 2018; Abedin et al., 2018a; Chen et al., 2018). Despite

DS being one of the key research topics in ﬁnancial decision support systems,

datasets with outliers have a signiﬁcant impact on the performance of DS classiﬁers.

Note that throughout this book chapter, modeling credit scoring data and credit

default risk prediction procedures to support decision-making activities refer to

ﬁnancial decision support systems (FDSSs) (Abedin et al., 2018b).

The existence of abnormal data, for instance, an outlier seriously affects the

accuracy of DS performance and other operations (Souiden et al., 2022). The outlier

represents the data points in which there are anomalies or errors that affect the data

analysis and modeling (Zhang et al., 2021). Outliers occur in numerous ways

including omitted variables, data errors, sampling errors, variable construction, and

nonnormality (Adams et al., 2019). These problems seriously affect DS models in

the ﬁeld of credit risk forecasting, customer churn prediction, facial recognition,

medical diagnosis, speech recognition, and web text classiﬁcation (Kim, 2017;Ma

et al., 2020; Moula et al., 2017; Shen et al., 2018; Xu et al., 2017; Kamishima et al.,

2018; Xiao et al., 2019). Therefore, outlier detection, that is, the action of detecting

patterns that are signiﬁcantly different from the data sample, is a vital challenge in

machine learning (ML) (Fernández et al., 2022). The outlier in credit cards deals

with the fraudulent of clients. Outliers in computer systems indicate fundamental

malicious activities. Industrial outlier represents system faults, and medical outliers

indicate underlying diseases (Wang & Mao, 2020).

As the outlier affects the performance of the DS models, it generates poor

ﬁnancial decisions in many organizations, including banks and other ﬁnancial

institutions. The ﬁnancial sector by nature is an intensively data-driven industry,

as it manages large quantities of client data. That is why FDSSs such as bankruptcy

data, credit data, etc., have the potential to contain outliers (Nyitrai & Miklos, 2019;

Zhang et al., 2021). Outliers in ﬁnancial decision-making may lead to invalid

inferences, model parameter biases, and poor volatility prediction (Granea &

Veiga, 2010). Therefore, the detection of outliers is an important concern for the

detection of rare objects in real application domains, such as in ﬁnance, materials

science, health, and industry (Ma et al., 2020). Outlier detection is a technique that

improves the efﬁciency of FDSSs and exhibits a huge difference from other ﬁnancial

decisions (Cai et al., 2020). Outlier detection intends to detect fraud and money

laundering of ﬁnancial decision systems by ﬁnding unusual customer behavior

patterns (Jun, 2006). Moreover, outlier detection in credit scoring domain helps to

reduce subjective elements in detecting outliers, eliminate the required time and

effort, and enhance the effectiveness of FDSSs (Okada et al., 2013; Yang et al.,

2022). Besides detection, outlier modiﬁcation should also be taken seriously,

because modiﬁcation helps preserve useful information at the time of modeling

FDSSs (Granea & Veiga, 2010).

Using Outlier Modiﬁcation Rule for Improvement of the Performance... 77

The presence and absence of outliers are the causes that differ the performance of

DS algorithms. More speciﬁcally, it is estimated that the performance of DS algo-

rithms may vary when there are 0%, 10%, and 20% outliers. The presence of outliers

can lead to destructive effects on the performance of DS algorithms if these are not

detected and modiﬁed precisely (Liu et al., 2021). Therefore, detection and modiﬁ-

cation of these outliers are the primary steps to generate more stability of DS

algorithms. This study applies simulated data to see the performance of DS algo-

rithms in the presence and absence of outliers before and after modiﬁcation. But

simulated data are often unable to reﬂect the present situation of corresponding

domains, which leads to unreliable and unrealistic reaction of people to the simula-

tion. That is why, to reduce prediction bias and enhance stability as well as

effectiveness of DS algorithms, the adaptation of real-life datasets (FDSS) is impor-

tant. For these reasons, this study uses credit scoring data as FDSS data.

In modeling FDSSs, this study analyzes the performance of four DS algorithms

such as LDA, k-NN, NB, and SVM by comparing the original and modiﬁed datasets.

Following the study by Nyitrai and Miklos (2019), this study trains multiple DS

classiﬁers to enhance the stability and minimize the forecast bias of the decision

support system. The modiﬁed dataset originates from the original data by applying

an outlier detection and modiﬁcation model based on Median (MED) and Median

Absolute Deviation (MAD). The current study applies both simulated and real-life

datasets to train the model. Real-life data refers to FDSSs data which come from

Credit Scoring Default Datasets. For measuring the performance of the DS algo-

rithms in absence (0) and presence of 10 and 20 outliers on original and modiﬁed

datasets, this study uses Accuracy, Sensitivity, Speciﬁcity, Detection rate (DR),

AUC, and pAUC. Our study makes notable contributions to DS performance and

FDSSs. This study extends the existing literature by comparing the performance of

DS algorithms on original and modiﬁed datasets. This paper informs stakeholders

that the detection and modiﬁcation of outliers is important to improve the perfor-

mance of DS algorithms and ﬁnancial decision-making. The current study suggests

that policymakers to motivate stakeholders to detect and modify outliers precisely

because an outlier-free dataset can result in a precise ﬁnancial decision. This study

also motivates ﬁnancial decision makers to improve the performance of applied DS

algorithms while making ﬁnancial and managerial decisions.

78 M. R. Auwul et al.

The paper proceeds as follows. In Sect. 2, we present a review of the related

literature. Section 3brieﬂy describes the proposed methodology along with applied

data science methods. The results and discussions are presented in Sect. 4. Finally,

Sect. 5concludes the paper with further road maps.

2 Related Literature

DS plays an important role in improving the performance of FDSS. Regarding

existing studies, Wang and Mao (2020) develop a dynamic ensemble outlier detec-

tion model to generate a base classiﬁer, determine the validation set, and estimate the

competence by using k-NN. Abedin et al. (2018a,b) utilize the topological applica-

tions of support vector machines (SVMs) and multilayer perceptrons (MLPs) to

conﬁrm the competitive performance of statistical intelligence mechanisms. Their

study deals with bankruptcy prediction and credit scoring in eight different databases

to assess FDSS. Li et al. (2022) offer a Fisher LDA classiﬁcation method attached

with Naïve Bayes (B-FLDA) for the event-related potential-based brain-computer

interface (ERP-BCI) to concurrently recognize the works, intentions, and idle states

of subject intentions.

Decision-making in banking and ﬁnance is now comparatively more complex

than in previous decades. One of the factors inﬂuencing ﬁnancial and banking

decisions is the existence of outliers. Leontitsis and Vorlow (2006) use the surrogate

data analysis (SDA) technique to deal with outliers which have an impact on stock

return. Their approach is based on the scale parameters of mean-stationary time

series and robust estimation of location. The study of Shen et al. (2018) determines

the effect of outliers on the relationship between ﬁnancial development and eco-

nomic growth. To conduct the study, they used a dynamic panel model by collecting

data from 48 countries between 1988 and 2014. To determine the effect of different

levels of outliers on the positive-valued insurance dataset, Okhli and Nooghabi

(2021) develop the contaminated exponential distribution as an alternative platform.

Detecting outliers is a vital phase in evaluating the impact of outliers in empirical

ﬁnance research. Adams et al. (2019) employ a multivariate identiﬁcation strategy to

identify and treat outliers appropriately in ﬁnancial data. To successfully detect the

ﬁnancial crisis, Domino (2020) introduces fourth-order multivariate cumulate

method as an outlier detection algorithm. Granea and Veiga (2010) applied a

wavelet-based general detection and correction method to detect isolated outliers

and outlier patches when modeling ﬁnancial time series data. Okada et al. (2013)

propose a case model to detect ﬁnancial outliers of the hospital industry, which helps

to reduce the required time and effort and enhance the quality of analysis. Based on

distance, Jun (2006) develops a cross-outlier detection model to detect outliers of

ﬁnancial transaction data. To minimize the negative impacts of outliers in the

þðÞ ð Þ

þ þ þðÞð Þ

ð Þ

noise-ﬁlled credit datasets, Zhang et al. (2021) propose a novel multistage ensemble

model with enhancing outlier adaption.

Using Outlier Modiﬁcation Rule for Improvement of the Performance... 79

Based on the literature cited above, the present study determines the following

research gaps. First, there is a range of studies dealing with the performance of DS

algorithms in the presence of outliers (Wang & Mao, 2020; Ling et al., 2020), but

none of them analyze the performance of DS algorithms by considering the absence

and presence of outliers, especially in FDSSs. Second, most studies apply MED or

MAD to detect outliers from datasets (Leys et al., 2013; Park & Moon, 2015; Abbas,

2019). That means previous studies are unable to demonstrate the performance of

outlier detection and modiﬁcation by combining both MED and MAD.

To ﬁll in the above research gaps, this study provides signiﬁcant theoretical

contributions to the existing literature on DS and FDSSs. First, this study extends

to previous studies by investigating the performance of DS algorithms by comparing

the original and modiﬁed datasets in the absence (0) and presence of 10% and 20%

outliers of FDSSs. Second, this paper combines MED and MAD as an outlier

detection and modiﬁcation algorithm in ﬁnancial decision-making.

3 Materials and Methods

To evaluate the performance of different DS algorithms for binary classiﬁcation, this

chapter applies the Receiving Operating Characteristics (ROC) curve, the area under

the ROC curve (AUC), and other classiﬁcation measures as follows:

Accuracy =TP þTNðÞ=TP þFP þTN þFNð,ð1Þ

Sensitivity =TP=TP FN , 2

Specificity =TN=TN FP , 3

Detection Rate =TP=TP FP TN FN , 4

MER =1–Accuracy, 5

where TP, FP, TN, and FN are the numbers of True Positive, False Positive, True

Negative, and False Negative, respectively. MER, AUC, and pAUC are the

misclassiﬁcation error rate, area under the ROC curve, and partial area under the

ROC curve, respectively.

On the basis of these parameters, this chapter declares a method as a good

performer if it produces larger values of Accuracy, Speciﬁcity, Sensitivity, Detection

Rate, AUC, and pAUC and low values of MER.

80 M. R. Auwul et al.

3.1 Statistical Methods to Be Compared

In this study, four popular classiﬁcation algorithms are evaluated, namely LDA, k-

NN, Naïve Bayes and SVM.

Linear Discriminant Analysis

LDA is a dimensionality reduction approach that is used as a pre-processing step to

classify patterns. LDA aims to design the features of higher dimensions of a space on

a lower dimension space to reduce resources and dimensional cost (Treder et al.,

2016). LDA represents a general discriminant function that uses a linear decision

boundary. For example, the target data instance zis classiﬁed by solving the

discriminant function d

for each K

class with the classiﬁcation rule S

. Let, the

prior probabilities is p(K

), the mean of each class is c

, and the common covariance

matrix is cov

. Then, the discriminant function is deﬁned as follows:

djzðÞ=log pK



-1

2cT

jcov-1

wcjþzTcov-1

wc:ð6Þ

The classiﬁcation rule for the target data instance is deﬁned as:

SkZðÞ=j:,j=arg max

|ﬄﬄﬄﬄ{zﬄﬄﬄﬄ}

djZðÞ:ð7Þ

k-Nearest Neighbors (k-NN)

k-NN is a widely used machine learning algorithm that is utilized in numerous

applications. k-NN is based on the assumption that the prediction value of the

example is probably the same as those of neighbors Jang et al. (2020). The k-NN

algorithm explains a metric in the predictor vector space, plots all applicants to a

position in this space, and evaluates posterior probability through the relative

amount of good risks between the k-nearest points in the training set.

Suppose Z

are the feature values, and K

denotes the labels of Z

for each j. Let the

number of classes be nand zbe the points for which the label is not known. To ﬁnd

the classes for unknown labels using k-NN, d(z,Z

), j=1, 2, ...,nﬁrst must be

determined for all values of k(dis a distance metric). Second, the distances are

determined for all n, the values are arranged in increasing order, and the distances are

taken from the sorted list (D≥0). Third, Dpoints are found that correspond to the

Ddistances. In the fourth step, let D

represent the number of data points belonging

to the jth class. In the ﬁfth step, put xin class iif D

,j≠i.8

Naïve Bayes (NB)

The NB classiﬁer is a probabilistic algorithm that is used for solving classiﬁcation

tasks based on the Bayes Theorem, where the independence of features is assumed.

The NB classiﬁer is widely applied in the data mining and product review sentiment

classiﬁcations domains (Xu et al., 2020).

Using Outlier Modiﬁcation Rule for Improvement of the Performance... 81

Let zbe a class variable that needs to be predicted and x

,...,x

are features,

then according to the Bayes Theorem, the probability of obtaining classes for zbased

on x’s is:

jx1,x2,...:,xn

ðÞ=Px

1jzðÞPx

2jzðÞ...Px

njzðÞPzðÞ

ðÞPx

ðÞ...Px

ðÞ :ð8Þ

As the denominator is unchangeable and the features are independent, the

denominator can be removed, and the result is proportionally given as:

jx1,x2,...:,xn

ðÞ1PzðÞ

i=1Px

ijzðÞ:ð9Þ

So, the class is obtained by ﬁnding the maximum probability as follows:

z=arg max

|ﬄﬄﬄﬄ{zﬄﬄﬄﬄ}

PzðÞ

i=1

ijzðÞ:ð10Þ

Support Vector Machine (SVM)

SVM refers to a machine learning model that is used to ﬁx pattern recognition

problems such as outlier detection, classiﬁcation, and regression. It utilizes the idea

of decision planes that apply decision boundaries to optimally distinct data into

numerous categories (Huang et al., 2021). The main objective of SVM is to ﬁnd the

hyper plane that classiﬁes the classes accurately with the maximum margin. The

linear SVM formula is given below. Suppose Xare the features and zare the target

values that need to be predicted. Then predict zas a function of the weighted values

of X. The Hinges loss function with a regularization term is deﬁned as:

Total cost =ωkk

2þK:ð11Þ

That is, the total cost is the sum of all losses for each observation. Here, ωdenotes

the weight value, and Kis the hyperparameter that controls the amount of regular-

ization. If Kis sufﬁciently small, this indicates a hard-margin classiﬁer, while for

large Kwe obtain a soft-margin classiﬁer.

3.2 Proposed Method

The current chapter proposes a novel methodology by combining MED and MAD as

an outlier mining (detection and modiﬁcation) method to evaluate the performance

of data analytics algorithms. This paper considers 0%, 10%, and 20% outliers to

assess how machine learning algorithms perform on original and modiﬁed datasets at

different levels of outlier existence. MED and MAD are the robust estimator of

location and robust measure of dispersion, respectively.

82 M. R. Auwul et al.

Yes

Data preprocessing and

normalizing

Financial ex-

pressed data

Checking the presence of outliers using

the proposed outlier detection method

Apply the classical

methods to identify class

labels on original data

Calculate validity

matrices

Apply the classical al-

gorithms to identify class

labels on MGE data

Fig. 1 Flow chart of the proposed outlier modiﬁcation algorithm

For this study, suppose that z

ijk

is the ith data instance for the jth that replicates in

the kth class and E(z

ijk

)=μ

, var(z

ijk

)=σ

represent the mean and variance value

for the ith data instance and the kth class (i=1, 2, ...,p;j=1, 2, ...,n

;k=1, 2, ...,

m), respectively. Based on this concept, we propose the following outlier

modiﬁcation rule:

(i) This chapter classiﬁes an expression of a sample as an outlier, if it does not fall

in the interval [MED

i,(k)

-L*NMAD

i,(k)

, MED

i,(k)

+L*NMAD

i,(k)

]. We declare

the corresponding sample as an outlying sample. Here, L=3 (for our study),

MED

i,(k)

=median(z

ij,(k)

); i=1, 2,...,p;j=1, 2, ...,n

;k=1, 2, ...,m) are

the median expressions of the ith data instance in the kth class, MAD

i,(-

=median

j=1,2,...,nk

(|z

ij,(k)

)-MED

i,(k)

|) is the median absolute deviation

and NMAD

i,(k)

=MAD

i,(k)

/0.6745 is the normalized MAD

i,(k)

of the ith

instance in the kth group.

(ii) For each sample from each group (k=1, 2, ...,m), check separately the

presence of outliers using Step 1. If an outlier is present, then replace it by the

median of the respective group [MED

i,(k)

], and get our desired modiﬁed

ﬁnancial expression (MFE) data.

(iii) Finally, apply the classical methods (DS Algorithms) in the MFE data to

identify the class label and ﬁnding different indices measurement such as

accuracy, sensitivity, speciﬁcity, detection rate, misclassiﬁcation error rate,

AUC and pAUC.

Gaussian noise

d +d

Using Outlier Modiﬁcation Rule for Improvement of the Performance... 83

The ﬂow chart of the proposed outlier modiﬁcation algorithm is depicted in

Fig. 1.

4 Results

This section illustrates the results of credit card fraud detection by using four DA

algorithms such as LDA, k-NN, NB, and SVM. All experiments were carried out on

a simulated dataset and three real-life credit card fraud datasets. Performance was

evaluated by comparing the original and modiﬁed datasets. This study utilized R

packages for these algorithms: class, caret, ROC, kkNN, e1071, and rpart. To judge

the performance of these algorithms, we used the MASS R package. The compre-

hensive R archive network (cran) or Bio-conductor are the main sources of these

packages. In this chapter, the terms “proposed”and “classical”refer to the applica-

tion of four traditional methods in the proposed and original MFED datasets,

respectively.

4.1 Simulated Data Analysis

Simulated data were generated for two groups (k=2) with known characteristics

both in the presence of 0%, 10%, and 20% outliers that mimic the nature of real-life

credit card data modeling scenarios. This study uses a data generation model that is

described in Table 1. In Table 1, the row represents the feature, and the column

represents the sample groups. For randomization, this study adds Gaussian noise to

the datasets. The generated data contains p=1,000 features consisting of two groups

(P1=P2=500) with sample size n=10. We set the value of the parameter das 0.2

and the noise parameter, σ

=0.05 to generate datasets for each of the data types.

This study generates 100 datasets from the data generating model as presented in

Table 1. The performance of four DA algorithms (LDA, k-NN, NB, and SVM) was

evaluated by comparing the original and modiﬁed datasets with two groups (k=2).

This study also evaluates the performance of these methods in the presence and

absence of outliers. To generate outlier datasets, this study randomly selects a dataset

containing 0%, 10%, and 20% outlier and replaces it with Gaussian noise with mean

60 and variance 3, respectively. This study measures different percentage of outlier

features (10% and 20%) with randomly choosing one or two samples. This study

computes different performance measures such as accuracy, sensitivity, speciﬁcity,

Table 1 Matrix used to gen-

erate simulation study Sample

S1S2S3

Group-1(p

)-a-d–a+d+d +N(0, σ

)

Group-2(p

)a-a-d

Data structure

detection rate, AUC, and pAUC for each of the 100 datasets using the seven DA

algorithms. Then, this paper determines the average of these performance measures

for each of the data types.

84 M. R. Auwul et al.

Table 2 Performance evaluation of four classiﬁers based on original and modiﬁed training dataset

for simulated data

Validity

matrices

Classical algorithms Proposed algorithms

LDA k-NN NB SVM LDA k-NN NB SVM

In absence of

outliers

Accuracy 0.977 0.941 0.977 0.968 0.977 0.941 0.977 0.968

Sensitivity 0.977 0.943 0.976 0.968 0.977 0.943 0.976 0.968

Speciﬁcity 0.977 0.938 0.979 0.968 0.977 0.938 0.979 0.968

Detection

rate

0.977 0.943 0.976 0.968 0.977 0.943 0.976 0.968

AUC 0.997 0.984 0.997 0.995 0.997 0.984 0.997 0.995

pAUC 0.198 0.186 0.198 0.195 0.198 0.186 0.198 0.195

In the presence

of 10% outliers

Accuracy 0.495 0.940 0.500 0.500 0.976 0.957 0.977 0.966

Sensitivity 0.573 0.938 0.550 0.550 0.976 0.961 0.975 0.968

Speciﬁcity 0.417 0.942 0.450 0.450 0.976 0.953 0.979 0.964

Detection

rate

0.573 0.938 0.550 0.550 0.976 0.961 0.975 0.968

AUC 0.516 0.982 0.724 0.598 0.997 0.966 0.997 0.994

pAUC 0.046 0.184 0.088 0.055 0.197 0.178 0.197 0.195

In the presence

of 20% outliers

Accuracy 0.500 0.928 0.500 0.500 0.976 0.946 0.976 0.962

Sensitivity 0.750 0.940 0.650 0.550 0.977 0.957 0.978 0.965

Speciﬁcity 0.250 0.915 0.350 0.450 0.974 0.936 0.973 0.959

Detection

rate

0.750 0.940 0.650 0.550 0.977 0.957 0.978 0.965

AUC 0.559 0.977 0.673 0.570 0.996 0.969 0.997 0.993

pAUC 0.070 0.197 0.048 0.175 0.194 0.062 0.179 0.197

For creating 100 Modiﬁed Financial Expressed Datasets (MFED), this chapter

ﬁrst applies the proposed outlier modiﬁcation technique for 100 training datasets.

The value of validity matrices such as accuracy, sensitivity, speciﬁcity, detection

rate, AUC, and pAUC are averaged over 100 datasets that are obtained from MFED

datasets. These average performance values are summarized in Table 2.We per-

ceived that in absence of outlier all four classiﬁers (LDA, k-NN, NB, and SVM)

produce same results using original data and proposed modiﬁed training dataset.

Nevertheless, in the presence of 10% and 20% outliers, the four classiﬁers performed

much better using modiﬁed training data than original training data. For instance, the

average accuracies 0.976, 0.957, 0.977, and 0.966 are produced by LDA, k-NN, NB,

and SVM, respectively, in the presence of outliers in each of 10% outliers that are

larger than 0.495, 0.940, 0.500, and 0.500, those were produced by the classical

classiﬁers in the same condition. The average accuracies 0.976, 0.946, 0.976, and

0.962 are produced by LDA, k-NN, NB, and SVM, respectively, in the presence of

outliers in each of 20% outliers that are larger than 0.500, 0.928, 0.500, and 0.500,

those were produced by the classical classiﬁers in the same condition. Hence, we

perceived that the performance of the classiﬁers improves by using MFED datasets

instead of the original datasets.

Using Outlier Modiﬁcation Rule for Improvement of the Performance... 85

0.1

0.2

0.3

0.4

0.5

0.6

LDA KNN NB SVM

Error Rate

In absence of outliers

10% outliers

Modified 10% outliers

20% outliers

Modified 20% outliers

Fig. 2 Performance evaluation using the average value of the error rate

The bar plot of the average value of error rate is presented in Fig. 2. From this

plot, this chapter determines that the error rate is approximately similar for both

classical and proposed algorithms in the absence of outliers (0%). But in case of 10%

and 20% outliers, error rate is raised for classical algorithms and the values are

getting lower for using MFED datasets.

In Fig. 3a, b, this study represented the box plot of the accuracies for 100 datasets

for 10% and 20% outlying datasets including original datasets for both classical and

proposed algorithms. Figure 3shows that for this simulation study, the performance

of the popular DS algorithms improves when the training datasets are modiﬁed by

the proposed method in the presence of outliers. Otherwise, these DS algorithms

produce the same results on original datasets.

4.2 Credit Card Default Data (CCDD)

To examine the performance of the four well-known DS algorithms (LDA, k-NN,

NB, and SVM), this study generated training and test datasets by randomly

partitioning (70% training and 30% test) the whole CCDD dataset into two inde-

pendent datasets. The log-transformed dataset was considered to remove unusual or

extreme values in this study. First, the training CCDD dataset was used in the

proposed outlier modiﬁcation procedure to obtain a modiﬁed training dataset as

described above. Thereafter, the performance of DS algorithms was determined

based on performance measures such as accuracy, sensitivity, speciﬁcity, detection

rate, and misclassiﬁcation error rate (MER) on CCDD datasets. Table 3shows the

average accuracy value using 100 simulations. The results indicate that all four DS

algorithms (LDA, k-NN, NB, and SVM) produce similar performance to those for

the original CCDD training dataset. On the contrary, these DS algorithms performed

far better on the modiﬁed CCDD datasets. For example, LDA produces accu-

racy =0.791 for the modiﬁed CCDD dataset, which is better than accuracy =0.768

using the original CCDD dataset. Figure 4a represents the boxplot of test values.

86 M. R. Auwul et al.

0.88 0.90 0.92 0.94 0.96 0.98 1.00

Accuracy

0.90 0.92 0.94 0.96 0.98 1.00

Accuracy

LDA oLDA mLDA KNN oKNN mKNN NB oNB mNB SVM oSVM mSVM

(a) Boxplot of test accuracies for 10% outliers case

(b) Boxplot of test accuracies for 20% outliers case

In absence of outliers

In presence of 10% outliers

Modified 10% simulated Data

In absence of outliers

In presence of 20% outliers

Modified 20% simulated Data

Fig. 3 Performance evaluation of four classiﬁers using boxplot (a) in presence of 10% outliers (b)

in presence of 20% outliers

Data Measure

Using Outlier Modiﬁcation Rule for Improvement of the Performance... 87

Table 3 Performance evaluation of four classiﬁers based on the original and modiﬁed training

dataset for real credit default datasets

Original data Modiﬁed data

LDA k-NN NB SVM LDA k-NN NB SVM

Default data Accuracy 0.768 0.769 0.767 0.768 0.791 0.770 0.769 0.776

Sensitivity 0.999 0.998 0.999 0.999 0.999 0.999 0.999 0.999

Speciﬁcity 0.001 0.002 0.001 0.053 0.062 0.413 0.018 0.001

Detection

rate

0.999 0.998 0.982 0.999 0.904 0.997 0.999 0.990

MER 0.232 0.231 0.233 0.232 0.209 0.230 0.231 0.224

Taiwan credit

default data

Accuracy 0.735 0.773 0.478 0.782 0.775 0.779 0.624 0.817

Sensitivity 0.919 0.990 0.459 0.990 0.990 0.999 0.579 0.959

Speciﬁcity 0.087 0.010 0.544 0.078 0.013 0.002 0.579 0.320

Detection

rate

0.918 0.990 0.489 0.990 0.990 0.999 0.637 0.959

MER 0.265 0.227 0.522 0.212 0.225 0.221 0.376 0.183

PAK credit

default data

Accuracy 0.739 0.738 0.715 0.738 0.739 0.739 0.738 0.739

Sensitivity 0.999 0.999 0.937 0.999 0.999 0.999 0.999 0.999

Speciﬁcity 0.001 0.001 0.087 0.001 0.001 0.001 0.001 0.001

Detection

rate

0.999 0.999 0.937 0.999 0.999 0.999 0.999 0.999

MER 0.261 0.262 0.285 0.262 0.261 0.261 0.262 0.261

4.3 Taiwan Credit Default Data

As in the same procedure as in the previous subsection, the entire Taiwan credit

dataset was divided into two independent datasets. To remove the unusual or

extreme values in this dataset, the log-transformed Taiwan dataset was considered

in this study. Firstly, the training Taiwan dataset was used in the proposed outlier

modiﬁcation procedure to obtain the modiﬁed training dataset as described above.

Thereafter, accuracy, sensitivity, speciﬁcity, detection rate, and MER were measured

using test Taiwan datasets. Table 3summarizes the average values of accuracy over

50 simulations. Table 3shows that all four classiﬁers (LDA, k-NN, NB, and SVM)

produce slightly better results using the modiﬁed Taiwan dataset than the original

one. For example, LDA produces an accuracy of 0.775 using the modiﬁed training

Taiwan dataset, which is greater than the accuracy of 0.735 using the original

training Taiwan credit dataset. Figure 4b represents the test accuracy values,

supporting the results in Table 3.

88 M. R. Auwul et al.

0.76 0.77 0.78 0.79

Accuracy

Original data

Modified CCDD Data

LDA mLDA KNN mKNN NB mNB SVM mSVM

(a) Boxplot of test accuracies for credit card default data

LDA mLDA KNN mKNN NB mNB SVM mSVM

Original data

Modified taiwan credit Data

Original data

Modified PAK credit Data

(b) Boxplot of test accuracies for taiwan credit default data

AccuracyAccuracy

0.7 0.8

0.6

0.5

0.74

0.73

0.72

0.71

Fig. 4 Performance evaluation of four classiﬁers using (a) CCDD dataset (b) Taiwan credit

dataset, and (c) PAK credit default dataset

Using Outlier Modiﬁcation Rule for Improvement of the Performance... 89

4.4 PAK Credit Default Data

Again, the whole PAK credit dataset was divided into two independent datasets, and

the log-transformed PAK credit dataset was used. Firstly, the PAK training credit

dataset was used in the proposed procedure to obtain the modiﬁed dataset. Thereaf-

ter, accuracy, sensitivity, speciﬁcity, detection rate, and MER were measured using

test PAK credit datasets. Table 3summarizes the average accuracies for 50 simula-

tions. From Table 3, notice that all four classiﬁers (LDA, k-NN, NB and SVM)

produce almost equal results using both the original PAK credit training dataset and

the modiﬁed PAK credit datasets except NB classiﬁers that gave better result for the

modiﬁed data than the original data. For example, NB produces accuracy =0.738

using the modiﬁed PAK credit dataset, which is greater than accuracy =0.715 using

the original training PAK credit dataset. The box plot of test accuracies is presented

in Fig. 4c.

Table 3summarizes the average values of the performance criteria estimated for

three well-known ﬁnancial datasets by different algorithms, respectively. We recon-

noiter similar interpretations like boxplots based on this table. We also perceived that

the proposed method produces almost parallel values of performance measures.

Therefore, we may conclude that the performance of the proposed algorithms

improved substantially over the performance of the classical algorithms.

5 Discussion

This is the ﬁrst study, as far as we know, that applies outlier mining-based data

analytics approaches in predicting credit card fraud. This chapter compares the

results and ﬁndings with some recently published papers. For example, Carcilloa

et al. (2021) apply hybrid unsupervised and supervised learning to detect credit card

fraud. Their results illustrate that the combined approach is more workable than the

baseline methods. Carneiro et al. (2017) develop a data mining-based methodology

to assess credit card default for an electronic merchant. They also state that a

combination of automatic and manual intelligent methodology offers feasible

insights. Vlasselaer et al. (2015) apply the data mining methodology and explain

that intrinsic and network-based features produce the most optimum results in

predicting credit card fraud customers. Bhattacharyya et al. (2011) also applied

data mining-based approaches to detect credit card fraud. They conclude that

traditional SVM, RF, and LR generate optimum prediction results than others. By

comparing and contrasting the results of other studies with ours, we can assert that

none of the existing studies covers outlier mining-based data analytic approaches in

predicting ﬁnancial status of credit card users as does this study.

90 M. R. Auwul et al.

6 Conclusion

One of the major objectives of DS algorithms is to extract knowledge from large

amount of data. In the literature, there exist many algorithms to perform this task.

However, it should be noted that most of them provide vague results in the presence

of outliers. Therefore, in this chapter, an outlier detection method and a modiﬁcation

rule were proposed to improve the classiﬁcation performance of several classiﬁca-

tion algorithms (LDA, k-NN, Naïve Bayes, and SVM). The performance of the

proposed methods was evaluated using both simulated and real ﬁnancial datasets.

The results indicate that all classiﬁcation algorithms produce misleading results in

the presence of outliers. However, their performance improved substantially when

using the proposed MFE data both for small and large datasets. From the data

analysis of the CCDD, Taiwan credit default, and PAK credit default tasks, we

conﬁrmed the effectiveness of the proposed method under real conditions.

References

Abbas, N. (2019). A robust S2 control chart with Tukey’s and MAD outlier detectors. Quality and

Reliability Engineering International, 36(1), 403–413. https://doi.org/10.1002/qre.2588

Abedin, M. Z., Guotai, C., Moula, F. E., Azad, A. S. M., & Khan, M. S. U. (2018a). Topological

applications of multilayer perceptrons and support vector machines in ﬁnancial decision support

systems. International Journal of Finance & Economics, 24(1), 474–507. https://doi.org/10.

1002/ijfe.1675

Abedin, M. Z., Guotai, C., Colombage, C., & Moula, F. E. (2018b). Credit default prediction using

a support vector machine and a probabilistic neural network. Journal of Credit Risk, 14(2),

1–27. https://doi.org/10.21314/JCR.2017.233

Abedin, M. Z., Guotai, C., Hajek, P., & Zhang, T. (2022). Combining weighted SMOTE with

ensemble learning for the class-imbalanced prediction of small business credit risk. Complex &

Intelligent Systems, 1–21.https://doi.org/10.1007/s40747-021-00614-4

Adams, J., Hayunga, D., Mansi, S., Reeb, D., & Verardi, V. (2019). Identifying and treating outliers

in ﬁnance. Financial Management, 48(2), 345. https://doi.org/10.1111/ma.12269

Bhattacharyya, S., Jha, S., Tharakunnel, K., & Westland, C. (2011). Data mining for credit card

fraud: A comparative study. Decision Support Systems, 50(3), 602–613. https://doi.org/10.

1016/j.dss.2010.08.008

Cai, S., Li, L., Li, S., Sun, R., & Yuan, G. (2020). An efﬁcient approach for outlier detection from

uncertain data streams based on maximal frequent patterns. Expert Systems with Applications,

160, 113646. https://doi.org/10.1016/j.eswa.2020.113646

Carcilloa, F., Borgnea, Y. L., Caelen, O., Kessaci, Y., Obleb, F., & Bontempia, G. (2021).

Combining unsupervised and supervised learning in credit card fraud detection. Information

Sciences, 557, 317–331. https://doi.org/10.1016/j.ins.2019.05.042

Carneiro, N., Figueira, G., & Costa, M. (2017). A data mining based system for credit-card fraud

detection in e-tail. Decision Support Systems, 95,91–101. https://doi.org/10.1016/j.dss.2017.

01.002

Chen, W., Zhang, S., Li, R., & Shahabi, H. (2018). Performance evaluation of the GIS-based data

mining techniques of best-ﬁrst decision tree, random forest, and naïve Bayes tree for landslide

susceptibility modeling. Science of the Total Environment, 644, 1006–1018. https://doi.org/10.

1016/j.scitotenv.2018.06.389

Using Outlier Modiﬁcation Rule for Improvement of the Performance... 91

Domino, K. (2020). Multivariate cumulants in outlier detection for ﬁnancial data analysis.

Physica A, 558, 124995. https://doi.org/10.1016/j.physa.2020.124995

Fernández, Á., Bella, J., & Dorronsoro, J. R. (2022). Supervised outlier detection for classiﬁcation

and regression. Neurocomputing, 486,77–92.

Granea, A., & Veiga, H. (2010). Wavelet-based detection of outliers in ﬁnancial time series.

Computational Statistics and Data Analysis, 54, 2580–2593. https://doi.org/10.1016/j.csda.

2009.12.010

Hajek, P., & Abedin, M. Z. (2020). A proﬁt function-maximizing inventory backorder prediction

system using big data analytics. IEEE Access, 8, 58982–58994. https://doi.org/10.1109/

ACCESS.2020.2983118

Huang, J., Jin, T., Liang, M., & Chen, H. (2021). Prediction of heat exchanger performance in

cryogenic oscillating ow conditions by support vector machine. Applied Thermal Engineering,

182, 116053. https://doi.org/10.1016/j.applthermaleng.2020.116053

Jang, S., Jang, Y. E., Kim, Y. J., & Yu, H. (2020). Input initialization for inversion of neural

networks using k-nearest neighbor approach. Information Science, 519, 229–242. https://doi.

org/10.1016/j.ins.2020.01.041

Jun T. (2006, April 9). A cross datasets referring outlier detection model applied to suspicious

ﬁnancial transaction discrimination. In Intelligence and Security Informatics, International

Workshop, WISI 2006, Proceedings, Vol. 3917, pp. 58–65. https://doi.org/10.1007/

11734628_7

Kamishima, T., Akaho, S., Asoh, H., & Sakuma, J. (2018). Model-based and actual independence

for fairness-aware classiﬁcation. Data Mining and Knowledge Discovery, 32, 258–286. https://

doi.org/10.1007/s10618-017-0534-x

Kim, K. (2017). Liquidity basis between credit default swaps and corporate bonds markets.

International Review of Economics & Finance, 48,98–115.

Kowalewski, O., & Pisany, P. (2022). Banks’consumer lending reaction to ﬁntech and bigtech

credit emergence in the context of soft versus hard credit information processing. International

Review of Financial Analysis, 81, 102116.

Kulczycki, P., & Franus, K. (2020). Outlier detection, clustering, and classiﬁcation –Methodolog-

ically united procedures for conditional approach. In N. T. Nguyen, B. H. Hoang, C. P. Huynh,

D. Hwang, B. Trawinski, & G. Vossen (Eds.), Computational collective intelligence. ICCCI

2020. Lecture notes in computer science (Vol. 12496). Springer. https://doi.org/10.1007/978-3-

030-63007-2_26

Leontitsis, A., & Vorlow, C. E. (2006). Accounting for outliers and calendar effects in surrogate

simulations of stock return sequences. Physica A, 368, 522–530. https://doi.org/10.1016/j.

physa.2005.12.037

Leys, C., Ley, C., Klein, O., Bernard, P., & Licata, L. (2013). Detecting outliers: Do not use

standard deviation around the mean, use absolute deviation around the median. Journal of

Experimental Social Psychology, 49, 764–766. https://doi.org/10.1016/j.jesp.2013.03.013

Li, X., Yang, S., Fan, R., Yu, X., & Chen, D. (2018). Discrimination of soft tissues using laser-

induced breakdown spectroscopy in combination with k nearest neighbors (kNN) and support

vector machine (SVM) classiﬁers. Optics and Laser Technology, 102, 233–239. https://doi.org/

10.1016/j.optlastec.2018.01.028

Li, M., Zhang, P., Yang, G., Xu, G., Guo, M., & Liao, W. (2022). A ﬁsher linear discriminant

analysis classiﬁer fused with naïve Bayes for simultaneous detection in an asynchronous brain-

computer interface. Journal of Neuroscience Methods, 371, 109496.

Ling, X., Yanlin, J., Jin, X., Xin, G., & Jing, H. (2020). GMDH-based outlier detection model in

classiﬁcation problems. Journal of Systems Science and Complexity, 33, 1516–1532. https://doi.

org/10.1007/s11424-020-9002-6

Liu, W., Fan, H., & Xia, M. (2021). Step-wise multi-grained augmented gradient boosting decision

trees for credit scoring. Engineering Applications of Artiﬁcial Intelligence, 97, 104036. https://

doi.org/10.1016/j.engappai.2020.104036

92 M. R. Auwul et al.

Ma, Y., Liu, S., & Li, Q. (2020). An advanced multiple outlier detection algorithm for 3D similarity

datum transformation. Measurement, 163, 107945.

Moula, F. E., Guotai, C., & Abedin, M. Z. (2017). Credit default prediction modeling: an

application of support vector machine. Risk Manag, 19, 158–187. https://doi.org/10.1057/

s41283-017-0016-x

Murinde, V., Rizopoulos, E., & Zachariadis, M. (2022). The impact of the FinTech revolution on

the future of banking: Opportunities and risks. International Review of Financial Analysis, 81,

102103.

Nyitrai, T., & Miklos, V. (2019). The effects of handling outliers on the performance of bankruptcy

prediction models. Socio-Economic Planning Sciences, 67,34–42. https://doi.org/10.1016/j.

seps.2018.08.004

Okada, S., Nagase, K., Ito, A., Ando, F., Nakagawa, Y., Okamoto, K., Kume, N., Takemura, T.,

Kuroda, T., & Yoshihara, H. (2013). Development of a methodology for the detection of

hospital ﬁnancial outliers using information systems. The International Journal of Health

Planning and Management, 29, e207–e232. https://doi.org/10.1002/hpm.2194

Okhli, K., & Nooghabi, M. J. (2021). On the contaminated exponential distribution: A theoretical

Bayesian approach for modeling positive-valued insurance claim data with outliers. Applied

Mathematics and Computation, 392, 125712. https://doi.org/10.1016/j.amc.2020.125712

Park, J. W., & Moon, Y. S. (2015). Robust estimation of target scale by removing outlier motion

vectors using MAD. Electronics Letters, 58(9), 691–693. https://doi.org/10.1049/el.2014.3821

Shen, C. H., Fan, X., Huang, D., Zhu, H., & Wu, M. W. (2018). Financial development and

economic growth: Do outliers matter? Emerging Markets Finance and Trade, 54(13), 2925–

2947.

Souiden, I., Omri, M. N., & Brahmi, Z. (2022). A survey of outlier detection in high dimensional

data streams. Computer Science Review, 44, 100463.

Treder, M. S., Porbadnigk, A. K., Avarvand, F. S., Muller, K. R., & Blankertz, B. (2016). The LDA

beamformer: Optimal estimation of ERP source time series using linear discriminant analysis.

NeuroImage, 279–291, 279. https://doi.org/10.1016/j.neuroimage.2016.01.019

Vlasselaer, V. V., Bravo, C., Caelen, O., Eliassi-Rad, T., Akoglu, L., Snoeck, M., & Baesens,

B. (2015). APATE: A novel approach for automated credit card transaction fraud detection

using network-based extensions. Decision Support Systems, 75,38–48. https://doi.org/10.1016/

j.dss.2015.04.013

Wang, B., & Mao, Z. (2020). A dynamic ensemble outlier detection model based on an adaptive

k-nearest neighbor rule. Information Fusion, 63,30–40. https://doi.org/10.1016/j.inus.2020.

05.00

Xiao, J., Tian, Y., Xie, L., Jiang, X., & Huang, J. (2019). A hybrid classiﬁcation framework based

on clustering. IEEE Transactions on Industrial Informatics, 16(4), 2177–2188. https://doi.org/

10.1109/TII.2019.2933675

Xu, P., Ding, Z., & Pan, M. (2017). An improved credit card users default prediction model based

on RIPPER. In In 2017 13th International conference on natural computation, fuzzy systems

and knowledge discovery (ICNCFSKD) (pp. 1785–1789). IEEE.

Xu, F., Pan, Z., & Xia, R. (2020). E-commerce product review sentiment classiﬁcation based on a

naïve Bayes continuous learning framework. Information Processing and Management, 57,

102221. https://doi.org/10.1016/j.ipm.2020.102221

Yang, F., Qiao, Y., Abedin, M. Z., & Huang, C. (2022). Privacy-preserved credit data sharing

integrating blockchain and federated learning for industrial 4.0. IEEE Transactions on Indus-

trial Informatics, 2022,1.https://doi.org/10.1109/TII.2022.3151917

Zhang, L., Su, H., Zio, E., Zhang, Z., Chi, L., Fan, L., Zhou, J., & Zhang, J. (2021). A data-driven

approach to anomaly detection and vulnerability dynamic analysis for large-scale integrated

energy systems. Energy Conversion and Management, 234, 113926.

Default Risk Prediction Based on Support

Vector Machine and Logit Support Vector

Machine

Fahmida-E-Moula, Nusrat Afrin Shilpa, Preity Shaha, Petr Hajek,

and Mohammad Zoynul Abedin

Abstract This chapter aims to predict the credit customer default risk. We propose a

machine learning algorithm such as Support Vector Machine and a hybrid default

risk prediction model such as Logistic Regression and Support Vector Machine

being known as LogitSVM (LSVM) to access the credit default risk. We apply three

real-world credit databases to validate the probability and value of the proposed risk

appraisal hybrid approaches. This chapter uses Type-I Error, Type-II Error, and Root

Mean Squared Error (RMSE) to evaluate the performance of the algorithms. Empir-

ical ﬁndings show that hybrid model experimentation (LogitSVM) maximizes

overall accuracy and minimizes RMSE, Type-I error, and Type-II error. This study

is useful for stakeholders to develop a wide variety of approaches to predict risk of

default of the credit customer.

Keywords Credit default prediction · Support vector machine · Logistic regression ·

Hybrid methodology

Fahmida-E-Moula

School of Economics and Management, Dalian University of Technology, Dalian, China

N. A. Shilpa · P. Shaha

Department of Finance and Banking, Hajee Mohammad Danesh Science and Technology

University, Dinajpur, Bangladesh

P. Hajek

Science and Research Centre, Faculty of Economics and Administration, University of

Pardubice, Pardubice, Czech Republic

e-mail: petr.hajek@upce.cz

M. Z. Abedin (✉)

Department of Finance, Performance and Marketing, Teesside University International Business

School, Teesside University, Middlesbrough, Tees Valley, UK

e-mail: m.abedin@tees.ac.uk

©The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

M. Z. Abedin, P. Hajek (eds.), Novel Financial Applications of Machine Learning

and Deep Learning, International Series in Operations Research & Management

Science 336, https://doi.org/10.1007/978-3-031-18552-6_6

94 Fahmida-E-Moula et al.

1 Introduction

Risk assessment is the most signiﬁcant and crucial concern in banking companies

and ﬁnancial institutions (Zhao et al., 2022; Hajek et al., 2022; Efat et al., 2022;

Abedin et al., 2021; Shajalal et al., 2021). It is the process of managing the

uncertainty related to risks. A sound risk assessment allows banks to plan a strong

decision-making procedure that reduces ﬁnancial losses. Three different types of risk

are causes of ﬁnancial uncertainties, such as market risk, credit risk, and operational

risk (Abedin et al., 2019a). Credit risk or default risk means the risk that a lender

takes when borrowers are unable to make the required payments on their debt

obligations. According to Abedin et al. (2022), credit risk involves 60% of the

total risk for the banking industries. Therefore, credit risk is considered as a wide-

ranging multifaceted crisis that aims to know credit the performance of the credit

customers and forecast their uncertainties. This ﬁnancial crisis drastically reduces the

proﬁt margin. As a result, many banks and other ﬁnancial institutions face complex-

ities, and some experience economic failure. However, one of the drastic end results

of the collapse is that the bank’s client or creditor bankruptcy is known as the credit

default. Credit Default Prediction (CDP) is essential for ﬁnancial institutions that aim

to decline future losses by eliminating the new credit proposal and estimating the

probable default risk. The accuracy of credit forecasting is fundamental to the

proﬁtability and efﬁciency of ﬁnancial institutions. Even a few positive adjustments

in the precision of credit risk prediction of potential applicants with credit default

will lessen a massive future loss for the ﬁnancial industry (Abedin et al., 2019b).

According to Vapnik (1995), the Support Vector Machine (SVM) is an extensive

applied algorithm for credit approval data classiﬁcation. SVM-based non-parametric

intelligent methods are more appropriate for default risk data classiﬁcation since

ﬁnancial data involve speciﬁc character such as non-linearity and neutrality of

covariance matrices between two groups of credit customers’class. However, the

standalone predictive algorithm cannot create the best credit risk accuracy for all

prediction problems. Therefore, there is a growing concentration that existing

applications of standalone learners may be further enhanced by utilizing blending

or hybrid methods. The hybrid forecasting model means the blend of traditional and

current artiﬁcial intelligence (AI) techniques, which signiﬁes improved forecasting

capacity than the application of a single classiﬁer (Chi et al., 2019). Additionally, the

hybrid learning system outperforms a standalone algorithm that provides better

accuracy and fewer prediction errors when employed in modeling credit approval

datasets (Moula et al., 2017). The purpose of this chapter is to predict the default risk

of the credit customer to minimize the burden of the applied credit risk prediction

classiﬁers. Therefore, this chapter proposes one machine learning algorithm such as

a Support Vector Machine and a hybrid default risk prediction model such as

Logistic Regression and Support Vector Machine known as LogitSVM (LSVM) to

access the credit default risk. Empirical ﬁndings show that experimentation with the

hybrid model (LogitSVM) minimizes the RMSE, Type-I error, and Type-II error and

maximizes overall accuracies. This study is useful for policymakers who have the

opportunity to inspect customer ﬁnancial practices that are able to increase their

future capability.

Default Risk Prediction Based on Support Vector Machine and Logit Support... 95

2 Literature Review

Researchers use many statistical classiﬁers to predict the default risk of credit

customers. For instance, multivariate adaptive regression splines (MARS) (Lee

et al., 2006), survival models (Luo et al., 2016), linear discriminant analysis

(LDA) (Lu et al., 2022), and fuzzy logistic regression analysis (Yang et al., 2022).

Jiashen You and Tomohiro Ando (2013) use a statistical model for the concurrent

estimation of hazard rate, risk-free interest rate, and loss given default, as well as the

credit risk dependency structure. However, there are difﬁculties with using these

statistical classiﬁers to predict credit approval data analysis. For instance, some

hypothesizes such as the multivariate normality hypothesizes for independent vari-

ables are usually violated in reality which makes these models hypothetically

unacceptable for an example set.

Researchers also used many machine learning classiﬁers to predict credit cus-

tomer default risk analysis. Boyacioglu et al. (2009) employed SVMs, three multi-

variate statistical methods, and four different neural network models to the problem

of forecasting bank credit failures. Huang et al. (2007) investigated that SVM-based

credit prediction approach can properly classify applications as either accepted or

rejected, reducing creditors’risk and interpreting future savings. Lee (2007), Kim

and Ahn (2012) and Shin et al. (2005) used SVMs to Korean credit risk approval

dataset and bankruptcy prediction. Ding et al. (2008), Hui and Sun (2006), and Xie

et al. (2011) utilized SVMs for the credit modeling of Chinese listed companies.

Experimenting with a Peruvian microﬁnance credit database, Blanco et al. (2013)

employed several intelligence credit risk assessment models based on the MLP

approach. However, the standalone analytical algorithm cannot create the best credit

risk accuracy for all prediction problems.

Therefore, nowadays corporate analysts and academic modelers have paid special

attention to hybridization along with the non-parametric approaches (Son et al.,

2016). In order to deal with the restrictions of statistical models and standalone

predictive algorithm and to generate the best credit risk accuracy for all forecasting

problems, SVM and LogitSVM-based default risk prediction models (hybrid

models) are proposed in the literature. SVM is a ﬂexible and intelligent method

that creates additive data connections with fewer predictors. LogitSVM (hybrid

model) increases credit risk discrimination ability by ensuring variety of prediction

assignments, model augmentation, and multifunctionality. Lin (2009)explores a

two-stage blending method of LR with BPN to Taiwanese banks’distress database

in the bankruptcy prediction domain. The hybrid model not only improves the

prediction power but also minimizes the misclassiﬁcation error. Besides, the hybrid

technique applied in this chapter solves the over ﬁtting concerns of other studies.

Consequently, it improves the ability to discriminate default risk.

96 Fahmida-E-Moula et al.

3 Methodology

3.1 Datasets

We focus on three credit datasets including “Credit Approval,”“German Credit”and

“Japanese Credit”to verify the probability and effectiveness of the proposed credit

risk assessment model. The “Credit Approval”data comes from Alyuda

NeuroIntelligence (http://www.alyuda.com). This database consists of 238 samples

of non-risky customers and 262 samples of risky customers. Each case includes

twelve ﬁnancial and non-ﬁnancial characteristics and one class attribute. The Ger-

man and Japanese credit databases come from the UCI (University of California,

Irvine) machine learning database repository. The “German credit”dataset consists

of 700 non-risky and 300 risky customers. Each credit customer seizes seven

numerical, thirteen categorical attributes, and one target variable. The “Japanese

credit”dataset includes a total of 690 instances having 307 non-default creditors and

383 default creditors. It has ﬁfteen attributes that include nine nominal variables, six

continuous variables, and one class attribute. This chapter applies three different

types of training scheme, 30%:70%, 50%:50%, and 70%:30%, respectively, to

determine the most optimal one.

3.2 Forecast Algorithms

Support Vector Machine

The SVM is suitable for a small sample, nonlinear, and high-dimensional data. Two

types of SVM are now accessible (i) Linear SVM and (ii) Kernel SVM. Linear SVM

acts as an extremely fast machine learning algorithm and performs an original

proprietary algorithm with a view to solve multiclass problems in large datasets.

Kernel-based SVM is used for nonlinear data classiﬁcation. In a nonlinear situation,

SVM mainly uses a kernel function to chart the preliminary data in the high-

dimensional factor to attain linear separability. Through this, it assists to solve the

issue of linear inseparability in the initial factor.

For a linear separable data set (x

;i=1, 2, ...,n), x2R

and y2R

, the

separation hyperplane is gained by maximizing the interval or solving the

corresponding convex quadratic programming problem:

ωTxþb=0, ð1Þ

where ωis a parameter vector, xand bare sample data and offset, respectively. The

corresponding classiﬁcation decision function is:



þðÞ ð Þ

þ Þ

iþ





Default Risk Prediction Based on Support Vector Machine and Logit Support... 97

fxðÞ=sgn ωTxþb:ð2Þ

For a linearly non-separable data set, each sample point presents a relaxation

variable to symbolize a non-negative measure of the misclassiﬁcation error. The

following optimization problem represents the linear-non-separable SVM:

min 1

2ω

2þCXn

i=1ξi,ð3Þ

s:t:yiωxib≥1-ξi,ξi≥0,i=1,2, ...,n, 4

where Cis the penalty factor that controls the association between accuracy and

generalization in the credit prediction training set.

Combining the kernel function with soft interval maximization principle, the

decision function of nonlinear SVM can be obtained by using the dual function

and Lagrange optimization algorithm, as follows:

fxðÞ=sgn Xn

i=1ai,yiKx,xðÞþb



,ð5Þ

where a

≥0 symbolizes the Lagrange multiplier and K(x,x

) represents the kernel

function, in agreement with the Mercer theorem.

To reduce computationally expensive calculations, the inner product is replaced

with kernel function K(x

). It converts the credit forecasting input data into a high-

dimensional feature space where the credit forecasting problems are separable and

hence increases the ability of the learning machine. Common forms of such kernel

functions include:

(a) The linear kernel, K(x

)=xT

ixj

(b) The sigmoid kernel, K(x

)=tanh(γxT

ixjr

)=γxTxjrd; and

(d) The radial basis kernel, K(x

)=exp. (-γ|| x

-x

)

As a ﬁnal SVM classiﬁer, this chapter obtains the decision function as follows:

YxðÞ=sgn Xn

i=0yiαiKx,xi

ðÞþb

,ð6Þ

where Y(x) represents the SVM decision function, sgn is the sign of the decision

parameter, K(x, x

) represents the kernel function, α

is the Lagrange multipliers, and

bis the bias of the model.

Logistic Regression

Logistic regression (LR) is a widely applied credit default prediction data modeling

method. The response variable of (LR), i.e., the outcome is binary (0, 1). Therefore,

researchers can employ it to clarify the relationship between the occurrence of an

incident of interest and a set of probable descriptive variables. In the circumstance of

þðÞ ð Þ

credit approval data modeling, the outcome links up to the borrowing loan perfor-

mance of a borrower during a speciﬁed period, usually twelve months.

98 Fahmida-E-Moula et al.

However, LR represents a valuable classiﬁer on the basis of two foundations in

the context of credit risk appraisal modeling. Firstly, in an LR, subsequent proba-

bilities are determined directly, which makes it more comprehensible than more

versatile “black box”techniques. Secondly, LR-based data classiﬁcation has

exposed it to make robust and better predictions in benchmarking studies for credit

risk assessment (Guo et al., 2016; Caigny et al., 2018). Therefore, LR can afﬁrm

more difﬁcult data classiﬁers in credit approval data modeling.

3.3 Performance Measures

Previous studies propose a number of appraisal performance measures to assess

forecasting methods in the ﬁeld of credit approval data analysis (Abedin et al., 2018).

The evaluation of the forecasting capacity of a classiﬁer is built from a confusion

matrix. This matrix is a special tabulation of correctly and incorrectly predicted

examples for each class. A confusion matrix for binary classiﬁcation is as stated in

Table 1, where tp refers to true positive, tn is true negative, fp means false positive

and fn represents false negative.

The Type-I error represents that a creditor with good status is misclassiﬁed as a

creditor with bad status in Eq. (7) and the Type-II error states that a creditor with a

bad status is misclassiﬁed as a creditor with a good status in Eq. (8):

Type I error =fn=tp þfnðÞ,ð7Þ

Type II error =fp=fp tn :8

The root mean square error (RMSE) is the average root square difference between

the estimated and actual values, that is:

RMSE =ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

1=NðÞ

i=1

θi-Pi

ðÞ

sð9Þ

where Nrefers to the total number of credit approval samples, θ

represents a binary

display for the actual execution of the status variable (0 if non-default, 1 if default),

and P

is the estimated probability of default. Credit scoring with large errors is

Table 1 Confusion matrix for a classiﬁcation problem

Predicted observations

Predicted positive Predicted negative

Actual observations Actual positive tp fp

Actual negative fn tn

weighted more deeply in Eq. (9) since the errors are squared before determining the

mean. Therefore, this performance indicator is efﬁcient in estimating huge surplus

deviations.

Default Risk Prediction Based on Support Vector Machine and Logit Support... 99

4 Results

4.1 Description of the Data

A summary of the three datasets is available in Table 2. The total number of

customers ranged from 500 to 1000, while two datasets were almost balanced (Credit

approval and Japanese credit datasets), and the German dataset was imbalanced in

favor of risky customers. The dimensionality of the databases ranged from 13 to 20.

4.2 Prediction of Credit Risk

From the experimental results shown in Tables 3and 4, for the “Credit Approval”

dataset, we can ﬁnd that the overall total accuracy of LSVM is 92.8%, while SVM is

92.7%. Moreover, it reveals that the overall LSVM generates the smallest RMSE and

Type–II error than the SVM. The total RMSE and Type–II error of LSVM are 5.35

and 1.98%, while SVM are 5.58 and 2.17%, respectively.

For the German credit dataset shown in Tables 5and 6, the total RMSE is the

same for both LSVM and SVM. The average Type–I error is 29.3% in LSVM, while

it is 30.0% for SVM. Regarding the kernel functions used, LSVM with linear and

polynomial kernel functions performed best, with high accuracy and low RMSE and

Type–I and Type–II errors.

The results for the Japanese credit approval database are presented in Tables 7

and 8. The results report that the overall total accuracy of the LSVM is 92.1% while

it is 90.2% for the SVM. Furthermore, the results expose that the total RMSE is

0.557, the Type–I error is 34.2%, and the Type–II error is 21.2% of the LSVM. The

total RMSE, Type–I error, and Type–II error of SVM are 0.572, 36.2%, and 22.4%,

respectively. It is clear that the errors of LSVM are smaller than those of SVM for the

Japanese credit database.

Table 2 Description of databases used in the experiments

Total cases Non-risky/risky customers No. of attributes

Credit approval 500 238/262 13

German credit 1000 700/300 20

Japanese credit 690 307/383 15

LSVM model RMSE

Tr-dataset Te dataset Overall Type-I Type-II

(%) LSVM model RMSE

Tr-dataset Overall Type-I Type-II

100 Fahmida-E-Moula et al.

Table 3 Blending LogitSVM performance for the “Credit Approval”database

ratio

(%)

Risk assessment accuracy (%) Error (%)

a a a

30:70 LSVM –1 (LinK) 86.67 85.43 85.80 0.1867 19.85 7.17

LSVM –2 (RbfK) 86.67 87.14 87.00 0.4315 19.43 4.61

LSVM –3 (PolK) 86.67 87.14 87.00 0.3737 19.43 4.61

LSVM –4 (SigK) 51.33 48.86 49.60 0.7025 52.59 47.83

50:50 LSVM –1 (LinK) 84.40 87.60 86.00 0.3735 19.78 6.76

LSVM –2 (RbfK) 85.60 88.40 87.00 0.3601 19.43 4.61

LSVM –3 (PolK) 85.60 88.40 87.00 0.3601 19.43 4.61

LSVM –4 (SigK) 50.80 50.80 50.80 0.7085 52.00 47.33

70:30 LSVM –1 (LinK) 84.86 80.00 83.40 0.4181 19.84 13.17

LSVM –2 (RbfK) 86.86 87.33 87.00 0.3732 19.43 4.61

LSVM –3 (PolK) 86.86 87.33 87.00 0.3592 19.43 4.61

LSVM –4 (SigK) 50.00 50.67 50.20 0.7071 52.56 47.72

Note:

Tr refers to in-sample instances, while Te refers to out-sample instances. The overall results

are the average outcomes of the Tr and Te instances

Table 4 SVM performance for the “Credit Approval”database

TS ratio

Risk assessment accuracy (%) Error (%)

dataset

30:70 SVM –1 (LinK) 71.33 92.22 85.92 0.4072 16.87 11.29

SVM –2 (RbfK) 63.33 93.37 84.31 0.4315 11.33 18.71

SVM –3 (PolK) 76.00 93.37 88.13 0.3737 15.89 7.53

SVM –4 (SigK) 50.00 51.30 50.91 0.7025 52.31 47.02

50:50 SVM –1 (LinK) 83.60 87.20 85.40 0.3814 20.00 8.00

SVM –2 (RbfK) 85.60 88.40 87.00 0.3601 19.44 4.61

SVM –3 (PolK) 85.60 88.40 87.00 0.3601 19.44 4.61

SVM –4 (SigK) 48.80 50.80 49.80 0.7085 52.36 47.56

70:30 SVM –1 (LinK) 86.86 81.33 85.20 0.3973 20.29 8.04

SVM –2 (RbfK) 88.00 84.00 86.80 0.3732 18.15 7.39

SVM –3 (PolK) 88.00 84.67 87.00 0.3771 19.44 4.61

SVM –4 (SigK) 50.00 50.00 50.00 0.7071 52.61 47.78

4.3 Comparative Analysis of Prediction Models

To observe more reliability of the ﬁndings of current experimental setups, this

chapter applies a non-parametric Wilcoxon signed-ranks (WSR) test, which sets

the signiﬁcance level at p=0.01/0.05 to attach the statistically signiﬁcant perfor-

mance differences among the LogitSVM-based credit risk assessment classiﬁers.

Moreover, the objective of the study is to establish that the proposed hybrid

LSVM model RMSE

TS ratio (%) LSVM model RMSE

Tr-dataset Overall Type-I Type-II

algorithms are reliable learners to distinguish the non-risky customers from their

risky counterparts. However, in the database, all credit assessment classiﬁers (Model

Z) are veriﬁed for signiﬁcant dissimilarity from the perfect classiﬁer (Model A). The

null hypothesis represents the overall accuracy of Model A/type-I error/type-II

error =the overall accuracy of Model Z/type-I error/type-II error, while the inverse

is the alternative hypothesis. The column “improvement”states the relative progress

of the average CRA accuracy (type-I error/type-II error) that model A achieves over

model Z. The results are summarized in Tables 9,10, and 11.

Default Risk Prediction Based on Support Vector Machine and Logit Support... 101

Table 5 Blended LogitSVM performance for the German credit database

TS ratio

(%)

Risk assessment accuracy (%) Error (%)

Tr-dataset Te dataset Overall Type-I Type-II

30:70 LSVM –1 (LinK) 73.00 77.14 75.90 0.4988 19.97 37.61

LSVM –2 (RbfK) 75.00 72.26 73.10 0.5132 26.05 34.69

LSVM –3 (PolK) 74.33 77.71 76.70 0.4893 19.92 35.13

LSVM –4 (SigK) 42.00 45.86 44.70 0.7487 30.00 70.16

50:50 LSVM –1 (LinK) 75.00 75.60 75.30 0.4970 20.75 38.39

LSVM –2 (RbfK) 70.20 72.00 71.10 0.5375 27.75 43.24

LSVM –3 (PolK) 74.80 76.60 75.70 0.4928 20.64 37.27

LSVM –4 (SigK) 34.00 50.20 42.10 0.7591 29.67 70.00

70:30 LSVM –1 (LinK) 76.43 75.33 76.10 0.4911 20.00 36.96

LSVM –2 (RbfK) 70.86 73.33 71.60 0.5281 27.51 39.73

LSVM –3 (PolK) 75.43 74.67 75.20 0.4995 20.92 38.46

LSVM –4 (SigK) 46.00 41.67 44.70 0.7493 30.00 70.16

Table 6 SVM performance for the German credit database

Risk assessment accuracy (%) Error (%)

dataset

30:70 SVM –1 (LinK) 70.33 77.14 75.10 0.5114 21.42 37.92

SVM –2 (RbfK) 72.00 70.86 71.20 0.5345 28.27 38.46

SVM –3 (PolK) 71.67 77.29 75.60 0.5045 21.21 36.54

SVM –4 (SigK) 42.00 70.00 61.60 0.6547 30.00 70.00

50:50 SVM –1 (LinK) 74.20 78.00 76.10 0.4885 20.41 36.20

SVM –2 (RbfK) 71.80 70.60 71.20 0.5366 28.45 36.36

SVM –3 (PolK) 74.00 77.80 75.90 0.4906 20.99 35.88

SVM –4 (SigK) 34.00 50.00 42.00 0.7648 30.00 70.00

70:30 SVM –1 (LinK) 74.43 74.67 74.50 0.5045 21.22 40.09

SVM –2 (RbfK) 72.57 70.00 71.80 0.5357 27.95 32.69

SVM –3 (PolK) 75.00 77.33 75.70 0.4764 20.44 37.45

SVM –4 (SigK) 46.00 42.00 44.80 0.7482 30.00 70.00

Table 3shows that for 30%:70%, 50%:%50% and 70%:30% TSs, LSVM-3 has

the highest averages in overall credit risk assessment (CRA) accuracies. For the

(%) LSVM model RMSE

Tr-dataset Overall Type-I Type-II

TS ratio (%) LSVM model RMSE

Tr-dataset Overall Type-I Type-II

German credit dataset, Table 5shows that LSVM-3 has the highest averages in

overall credit risk assessment (CRA) accuracies in 30%:70% and 50%:%50% TSs,

but LSVM-1 has the highest accuracies for 70%:30% TSs. For the Japanese credit

dataset, Table 7represents that LSVM-3 has the highest accuracies in 30%:70% and

70%:30% TSs, but LSVM-2 has the highest accuracies in 50%:%50% TSs.

102 Fahmida-E-Moula et al.

Table 7 Blended LogitSVM performance for the Japanese credit database

TS ratio

Risk assessment accuracy (%) Error (%)

dataset

30:70 LSVM –1 (LinK) 76.81 84.68 82.32 0.4364 14.28 19.72

LSVM –2 (RbfK) 82.13 86.96 85.51 0.3920 21.33 6.99

LSVM –3 (PolK) 82.13 86.96 85.51 0.3920 21.33 6.99

LSVM –4 (SigK) 48.79 50.10 49.71 0.7110 55.49 44.48

50:50 LSVM –1 (LinK) 82.90 87.25 85.07 0.3853 19.82 10.23

LSVM –2 (RbfK) 83.48 87.54 85.51 0.3797 21.33 6.99

LSVM –3 (PolK) 83.48 87.25 85.36 0.3818 21.55 7.01

LSVM –4 (SigK) 53.04 52.75 52.90 0.6863 52.68 41.81

70:30 LSVM –1 (LinK) 83.02 89.37 84.93 0.3690 18.96 11.57

LSVM –2 (RbfK) 83.64 89.86 85.51 0.3614 21.33 6.99

LSVM –3 (PolK) 83.64 90.34 85.65 0.3576 21.11 6.97

LSVM –4 (SigK) 51.76 54.59 52.61 0.7175 53.01 42.18

Table 8 SVM performance for the Japanese credit database

Risk assessment accuracy (%) Error (%)

dataset

30:70 SVM –1 (LinK) 77.78 78.47 78.26 0.4677 28.84 13.48

SVM –2 (RbfK) 78.26 86.96 84.35 0.4341 20.48 11.05

SVM –3 (PolK) 82.13 86.96 85.51 0.3920 21.33 6.99

SVM –4 (SigK) 47.83 52.17 50.87 0.7070 55.93 44.76

50:50 SVM –1 (LinK) 75.07 86.67 80.87 0.4322 16.88 21.08

SVM –2 (RbfK) 83.48 87.54 85.51 0.3798 21.33 6.99

SVM –3 (PolK) 83.48 87.25 85.36 0.3818 21.39 7.27

SVM –4 (SigK) 51.01 47.83 49.42 0.7111 55.53 44.52

70:30 SVM –1 (LinK) 82.82 85.99 83.77 0.3944 22.22 10.03

SVM –2 (RbfK) 83.85 89.37 85.51 0.3639 21.33 6.99

SVM –3 (PolK) 83.85 89.37 85.51 0.3639 21.33 6.99

SVM –4 (SigK) 45.76 51.21 47.39 0.6951 55.41 44.19

Evidence from Tables 9–11 shows that in 30%:70% and 50%:50% TSs, LSVM-3

on the German credit database obtain a remarkable improvement compared to other

classiﬁers considering the overall CRA accuracy criterion. For type-I error, LSVM-3

yields more than 30% improvement for the same dataset in 50%:50%, while for

type-II error, LSVM-3 on a similar database attains more than 46% improvement. It

Model A Model Z Impr. (%) pImpr. (%) pImpr. (%) p

Default Risk Prediction Based on Support Vector Machine and Logit Support... 103

Table 9 Results of Wilcoxon signed-ranks test for the “Credit Approval”database

Ratio

(%)

Overall accuracy Type-I Error Type-II Error

30:70 LSVM-3 LSVM-1 1.3980 0.800 2.1159 1.02E-18

35.7043 5.97E-13

LSVM-2 0.0000 0.502 0.0000 1.62E-25

0.0000 1.31E-5

LSVM-4 75.4032 5.78E-7

63.0538 4.53E-55

90.3617 8.66E-28

50:50 LSVM-3 LSVM-1 1.1628 0.525 1.8013 1.44E-16

31.8147 8.83E-38

LSVM-2 0.0000 0.001

0.0000 1.29E-77

0.0000 9.49E-20

LSVM-4 71.2598 4.03E-7

62.6346 8.05E-32

90.2599 2.11E-37

70:30 LSVM-3 LSVM-1 4.3165 2.69E-6

2.0665 0.044

64.9962 6.09E-16

LSVM-2 0.0000 0.638 0.0000 2.17E-11

0.0000 7.09E-30

LSVM-4 73.3068 1.15E-4

63.0327 7.21E-83

90.3395 3.87E-43

α=0.01,

α=0.05

Table 10 Results of Wilcoxon signed-ranks test for the German credit database

Ratio

(%)

Overall accuracy Type-I Error Type-II Error

30:70 LSVM-3 LSVM-1 1.0540 1.81E-192

2.1159 4.47E-7

6.5940 4.36E-14

LSVM-2 4.9248 0.744 0.0000 5.58E-98

-1.2684 0.098

LSVM-4 71.5884 2.92E-22

63.0538 2.07E-17

49.9287 2.21E-33

50:50 LSVM-3 LSVM-1 0.5312 0.841 0.5329 1.34E-53

2.9174 1.07E-4

LSVM-2 6.4698 1.91E-17

25.6216 2.15E-101

13.8067 3.52E-10

LSVM-4 79.8100 4.37E-11

30.4348 4.29E-31

46.7571 8.63E-32

70:30 LSVM-3 LSVM-1 6.2849 2.04E-16

27.2992 0.953 6.9721 0.115

LSVM-2 1.1968 0.453 4.3977 0.072 3.9002 3.78E-24

LSVM-4 70.2461 1.40E-26

33.3333 9.11E-39

47.3204 7.64E-31

α=0.01

Table 11 Results of Wilcoxon signed-ranks test for the Japanese credit database

Ratio

(%)

Overall accuracy Type-I Error Type-II Error

30:70 LSVM-3 LSVM-1 3.8751 9.78E-186 -49.370 6.07E-7

64.5538 6.12E-5

LSVM-2 0.0000 0.451 0.0000 1.64E-8

0.0000 2.53E-23

LSVM-4 72.0177 2.09E-14

61.561 3.03E-31

84.2851 1.90E-7

50:50 LSVM-3 LSVM-1 0.5172 0.743 -7.6186 2.69E-25

31.6716 8.59E-60

LSVM-2 0.1757 0.421 1.0209 1.84E-61

0.2853 4.75E-38

LSVM-4 61.6446 1.96E-4

59.510 0.007

83.2815 3.17E-73

70:30 LSVM-3 LSVM-1 0.8478 0.027

-11.340 9.32E-5

39.7580 1.07E-36

LSVM-2 0.1637 0.003

1.0314 1.14E-13

0.2861 1.44E-71

LSVM-4 62.8017 0.344 60.177 4.38E-85

-83.4756 1.30E-83

α=0.01,

α=0.05

is clear from Tables 9–11 that all improvements in type-I error and type-II error on

all databases are statistically signiﬁcant with respect to the best-performing blending

classiﬁers. On the contrary, in some cases, the improvements of the accuracy

criterion regarding the best algorithms are statistically insigniﬁcant, and this is

mentioned in the fact that the best algorithms have spaces for further improvements

relative to their competing learners.

104 Fahmida-E-Moula et al.

5 Discussion

Jiashen You and Tomohiro Ando (2013) show that their numerical results verify the

practicality of their proposed statistical methodology. The empirical ﬁndings of

Boyacioglu et al. (2009) show that, as learning algorithms, SVMs with some neural

network architectures outperform the multivariate statistical methods. The ﬁndings

of Blanco et al. (2013) reveal that neural models outperform statistical techniques.

SVMs are the better approach to learn a small size of data patterns as opposed to

common DA, LR, and MLP (Kim & Ahn, 2012; Shin et al., 2005). On the other

hand, the result of Lin (2009) claims that the hybrid methodology outperforms the

baseline models by generating 80.8% prediction accuracy, while the baseline LR and

BPN provide 75.6% and 75.34%, respectively. Therefore, in this chapter, we discuss

SVM and LogitSVM (hybrid model), which are better than the performance of other

statistical methods and baseline models.

6 Conclusion

Credit default risk prediction is important to survive for both ﬁnancial and

non-ﬁnancial companies. Since the recent global ﬁnancial crisis has exposed, insuf-

ﬁcient decision-making not only affects proﬁtability but also threatens ﬁrm solvency

in the credit approval procedure. As a result, the accuracy of credit forecasting is

essential for the proﬁtability and solvency of ﬁnancial institutions. This study pre-

sents SVM and LogitSVM as new blended intelligent algorithms to assess credit

risk. We evaluate the performance of the algorithms using Type–I error, Type–II

error, and Root Mean Squared Error (RMSE). The results demonstrate that the

experimentation with the hybrid model (LogitSVM) minimizes the RMSE, Type–I

error, and Type–II error.

The present methodology is extensively applicable in many previous works.

Therefore, as a further avenue, further study will improve the investigated technique

utilizing more advanced algorithms. We would like to expand the current study as a

future line of research by including credit approval databases from other regions.

Moreover, the ﬁndings of this chapter relate to empirical approaches. Therefore,

future work may be further veriﬁed by applying a real-life case study.

Default Risk Prediction Based on Support Vector Machine and Logit Support... 105

References

Abedin, M. Z., Guotai, C., Colombage, S., & Moula, F. E. (2018). Credit default prediction by

using a support vector machine and a probabilistic neural network. Journal of Credit Risk, 14(2),

1–27.

Abedin, M. Z., Guotai, C., Moula, F. E., Zhang, T., & Hassan, M. K. (2019a). An optimized support

vector machine intelligent technique using optimized feature selection methods: Evidence from

Chinese credit approval data. Journal of Risk Model Validation, 13(2), 1–46.

Abedin, M. Z., Guotai, C., Moula, F. E., Azad, A. S. M. S., & Khan, M. S. U. (2019b). Topological

applications of multilayer perceptrons and support vector machine in ﬁnancial decision support

systems. International Journal of Finance & Economics, 24, 474–507.

Abedin, M. Z., Hasan, M. M., Hassan, M. K., & Hajek, P. (2021). Deep learning-based exchange

rate prediction during the COVID–19. Annals of Operations Research. https://doi.org/10.1007/

s10479-021-04420-6

Abedin, M. Z., Guotai, C., Hajek, P., & Tong, Z. (2022). Combining weighted smote with ensemble

learning for class-imbalanced prediction of small business credit risk. Complex & Intelligent

Systems.

Blanco, A., Mejías, R. P., Lara, J., & Rayo, S. (2013). Credit scoring models for the microﬁnance

industry using neural networks: Evidence from Peru. Expert Systems with Applications, 40,

356–364.

Boyacioglu, M. A., Kara, Y., & Baykan, O. K. (2009). Predicting bank ﬁnancial failures using

neural networks, support vector machines and multivariate statistical methods: A comparative

analysis in the sample of savings deposit insurance fund (SDIF) transferred banks in Turkey.

Expert Systems with Applications, 36, 3355–3366.

Chi, G., Uddin, M. S., Abedin, M. Z., & Yuan, K. (2019). Hybrid model for credit risk prediction:

An application of neural network approaches. International Journal on Artiﬁcial Intelligence

Tools, 28(5), 1–33. https://doi.org/10.1142/S0218213019500179

Ding, Y., Song, X., & Zen, Y. (2008). Forecasting ﬁnancial condition of Chinese listed companies

based on support vector machine. Expert Systems with Applications, 34, 3081–3089.

Efat, M. I. A., Hajek, P., Abedin, M. Z., Azad, R. U., Jaber, M. A., Aditya, S., & Hassan, M. K.

(2022). Deep-learning model using hybrid adaptive trend estimated series for modelling and

forecasting sales. Annals of Operations Research. https://doi.org/10.1007/s10479-022-04838-6

Guo, Y., Zhou, W., Luo, C., Liu, C., & Xiong, H. (2016). Instance-based credit risk assessment for

investment decisions in P2P lending. European Journal of Operational Research, 249,

417–426.

Hajek, P., Abedin, M. Z., & Sivarajah, S. (2022). Fraud detection in mobile money transactions

using an XGBoost-based framework. Information Systems Frontiers. https://doi.org/10.1007/

s10796-022-10346-6

Huang, C. L., Chen, M. C., & Wang, C. J. (2007). Credit scoring with a data mining approach based

on support vector machines. Expert Systems with Applications, 33, 847–856.

Hui, X., & Sun, J. (2006). An application of support vector machine to companies’ﬁnancial distress

prediction. Lecture Notes in Artiﬁcial Intelligence, 3885, 274–282.

Jiashen You & Tomohiro Ando. (2013). A statistical modeling methodology for the analysis of term

structure of credit risk and its dependency. Expert Systems with Applications, 40(12),

4897–4905.

Kim, K. J., & Ahn, H. (2012). A corporate credit rating model using multi-class support vector

machines with an ordinal pairwise partitioning approach. Computers & Operations Research,

39, 1800–1811.

Lee, Y. C. (2007). Application of support vector machines to corporate credit rating prediction.

Expert Systems with Applications, 33(1), 67–74.

Lee, T. S., Chiu, C. C., Chou, Y. C., & Lu, C. J. (2006). Mining the customer credit using

classiﬁcation and regression tree and multivariate adaptive regression splines. Computational

Statistics & Data Analysis, 50(4), 1113–1130.

106 Fahmida-E-Moula et al.

Lin, S. L. (2009). A new two-stage hybrid approach of credit risk in banking industry. Expert

Systems with Applications, 36, 8333–8341.

Lu, Y., Yang, L., Shi, B., Li, J., & Abedin, M. Z. (2022). A Novel Framework of Credit Risk

Feature Selection for SMEs during Industry 4.0. Annals of Operations Research. https://doi.org/

10.1007/s10479-022-04849-3

Luo, S., Kong, X., & Nie, T. (2016). Spline based survival model for credit risk modeling.

European Journal of Operational Research, 253(3), 869–879.

Moula, E. F., Goutai, C., & Abedin, M. Z. (2017). Credit default prediction modeling: An

application of support vector machine. Risk Management, 19(2), 158–187.

Shajalal, M., Petr, H., & Abedin, M. Z. (2021). Product backorder prediction with deep neural

network on imbalance data. International Journal of Production Research. https://doi.org/10.

1080/00207543.2021.1901153

Shin, K. S., Lee, T. S., & Kim, H. J. (2005). An application of support vector machines in

bankruptcy prediction model. Expert Systems with Applications, 28(1), 127–135.

Son, Y., Byun, H., & Lee, J. (2016). Nonparametric machine learning models for predicting the

credit default swaps: An empirical study. Expert Systems with Applications, 58, 210–220.

Vapnik, V. N. (1995). The nature of statistical learning theory. Springer.

Xie, C., Luo, C., & Yu, X. (2011). Financial distress prediction on SVM and MDA methods: The

case of Chinese listed companies. Quality & Quantity, 45, 671–686.

Yang, F., Qiao, Y., & Abedin, M. Z. (2022). Privacy-preserved credit data sharing integrating

blockchain and federated learning for industrial 4.0. IEEE Transactions on Industrial

Informatics.

Zhao, Y., Goodell, J. W., Dong, Q., Wang, Y., & Abedin, M. Z. (2022). Overcoming spatial

stratiﬁcation of ﬁntech inclusion: Inferences from across Chinese provinces to guide policy

makers. International Review of Financial Analysis, 84, 102411. https://doi.org/10.1016/j.irfa.

2022.102411

107

Predicting Corporate Failure Using

Ensemble Extreme Learning Machine

David Veganzones

Abstract Corporate failure prediction has become a major topic in the accounting

and ﬁnance literature. Effective prediction models are essential for banks and

ﬁnancial institutions to solve ﬁnancial decision-making problems. In general, artiﬁ-

cial intelligence and machine learning techniques have been mainly employed to

develop corporate failure models due to their prediction superiority in comparison to

the traditional statistical method. Extreme learning machine is a newly developed

artiﬁcial intelligence technique with an extremely fast learning speed. Nonetheless,

its performance instability may be a major constraint for its practical application. The

literature documents that the ensemble is one of the widely used methods to improve

the generalization performance of weak classiﬁers. Therefore, we propose in this

study an ensemble of extreme learning machine for improving the prediction

performance on corporate failure task. In particular, we compare four benchmark

ensemble methods (multiple classiﬁers, bagging, boosting, and random subspace) to

evaluate which is best suited for extreme learning machine. Experimental results on

French ﬁrms indicated that bagged and boosted extreme learning machine showed

the best-improved performance.

Keywords Forecasting · Corporate failure · Machine learning · Extreme learning

machine · Ensemble

1 Introduction

The global economic developments of recent decades have put corporate failure and

their consequences for economic well-being under the spotlight, to the extent that

bankruptcy or business failure has become a crucial task in ﬁnance. This, in turn, has

emphasized that ﬁnancial institutions need effective prediction mechanisms in order

to make an appropriate lending decision.

D. Veganzones (✉)

ESCE International Business School, OMNES Education, Paris La Défense, France

©The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

M. Z. Abedin, P. Hajek (eds.), Novel Financial Applications of Machine Learning

and Deep Learning, International Series in Operations Research & Management

Science 336, https://doi.org/10.1007/978-3-031-18552-6_7

108 D. Veganzones

In general, the objective of corporate failure prediction is to forecast the likeli-

hood that a ﬁrm will survive or fail with the minimum possible classiﬁcation error.

That is why corporate failure research aims at binary classiﬁcation (Séverin &

Veganzones, 2021; Ouenniche & Tone, 2017). From the binary classiﬁcation point

of view, the model’s output is a dichotomous variable that takes the value of 1 when

the ﬁrm follows a bankruptcy procedure and is set to 0 when the ﬁrm survives. The

explanatory variables to design corporate failure prediction models are often ﬁnan-

cial ratios, which measure the relationship between any two items on ﬁnancial

statements.

Since the pioneer studies of Beaver (1966) and Altman (1968) who documented

the predictive power of ratio analysis, many prediction techniques have been

employed to develop corporate failure prediction models, including statistical and

artiﬁcial intelligence methods (Veganzones & Severin, 2020; Kumar & Ravi, 2007;

Moula et al., 2017). On the one hand, researchers still employ well-known statistical

methods, notably linear discriminant analysis and logistic regression, due to their

simplicity and capacity to interpret the data, even though they are clearly

outperformed by machine learning techniques. On the other hand, artiﬁcial intelli-

gence techniques (i.e., support vector machine, decision trees, neural networks,

fuzzy set theory, self-organizing map) have become indispensable tools in the ﬁeld

of corporate failure prediction, especially in this era of advanced informatics and

computing technology (Abedin et al., 2021). Their superiority relies on the fact that

they learn directly from the data, which makes it possible to test complex data using

nonlinear approaches, and therefore, their predictions are more reliable. Nonetheless,

these mentioned methods are not free of drawbacks: low learning rate, slow com-

putational time, converge in local minima, etc. (Yu et al., 2014; Abedin et al., 2018),

which could make corporate failure prediction time consuming and arduous.

To overcome these, we consider a novel prediction method, Extreme Learning

Machine (ELM) (Huang et al., 2006a) to predict corporate failure. There are several

reasons behind choosing ELM as the classiﬁer for the prediction of corporate

failures. Firstly, despite many existing methodologies for predicting corporate fail-

ure, new methods of research should be continually explored by researchers and

practitioners. Secondly, the main concept behind ELM is the random initialization of

the Single Layer Feed-Forward Neural Network (SLFN), which replaces the com-

putationally cost procedure of training the hidden layer performed by other artiﬁcial

intelligence techniques. Unlike the AI techniques, it does not need to calibrate

parameters, such as the learning rate. For this reason, ELM has good performance

with an extremely fast learning speed (Akusok et al., 2015) and it is proven to be a

universal approximator given enough hidden neurons (Huang et al., 2006b).

However, as other techniques, ELM possesses a main drawback: the random

initialization that allows ELM to be an extremely fast algorithm, it becomes ELM a

highly unstable classiﬁer as well. In ELM, even if we train the same training sample

several times, it performs differently due to the random initialization of bias and

weights between the input and hidden nodes. Although the reliance on a single ELM

may be misguided, the ensemble of predictions might improve the generalization

performance of the ELM. Indeed, ensemble methods are usually used as an

instrument for improving the accuracy of the learning algorithm by constructing and

combining a set of weak classiﬁers (Kim & Kang, 2010; Abedin et al., 2022). This

rationale motivates our speciﬁc study of the performance of the ensemble extreme

learning machine to predict corporate failure.

Predicting Corporate Failure Using Ensemble Extreme Learning Machine 109

Consequently, the aim of this current work is to fully examine which is the best

ensemble procedure to improve the performance of ELM for corporate failure

prediction. This is of signiﬁcant importance because the diversity generation method

is key in the process of creating an ensemble of classiﬁers. According to Rokach

(2010), diversity creation can be obtained in several ways: by manipulating the

training sample, by manipulating the inducer, by varying the representation of the

target attribute and by changing the search space. Of all possible ensemble tech-

niques, we selected 4 based on their popularity in the literature (Verikas et al., 2010):

Multiple classiﬁers, Bagging, Boosting, and Random Subspace. The fact that the

chosen techniques rely on different ensemble procedures might provide further

insight into the general characteristics of ensemble techniques that are inﬂuenced

by the base classiﬁer. In turn, a rigorous study of such methods would provide

assistance in designing a model of corporate failure based on ensemble ELM.

Furthermore, optimal performance of prediction models developed based on ensem-

ble ELM models can be employed as a baseline prediction model for future research.

The rest of the paper is organized as follows. Section 2presents the research

methodology. Sections 3and 4describe the experimental design and results, respec-

tively. Finally, in Sect. 5, the conclusions are summarized.

2 Research Methodology

In this section, we present the method employed in this study. In particular, we

describe the extreme learning machine classiﬁer as well as the ensemble modeling

techniques.

2.1 Extreme Learning Machine

The Extreme Learning Machine (ELM) classiﬁer was proposed by Huang et al.

(2006a). The ELM represents a fast way of creating a Single Layer Hidden Feed-

Forward Neural Network (SLFN) by the random initialization of the internal bias

and weights. The hidden layer does not need to be iteratively tuned; it bypasses the

time-consuming calibration setup performed by artiﬁcial intelligence algorithms. As

a result, ELM is an extremely fast learning speed while being a simple method. The

ELM algorithm can be described as follows:

Consider a set of Nobservations with features x

2ℝ

and the corresponding

output labels Y2{-1, 1}

Nxc

. A SLFN with mneurons in the hidden layer is written

by the following sum:



110 D. Veganzones

Fig. 1 Architecture of the

multiple classiﬁer

Σm

j=1βjϕwjxiþbj=Yik,i=1, ...,Nk=1, ...,c,ð1Þ

where β

are the output weights, ϕis the activation function, w

are the input weights

and b

represents the biases. The Eq. (1) can be expressed in the form of a matrix as

Hβ=Y, where

ϕw1x1þb1

ðÞ⋯ϕwmx1þbm

ðÞ

⋮⋱⋮

ϕw1xNþb1

ðÞ⋯ϕwmxNþbm

ðÞ

A:ð2Þ

β=β1⋯βm

ðÞ

cY=Y1⋯YN

ðÞ

Then, the output weights βcan be calculated by the Ordinary Least Squares

method using the Moore-Penrose pseudo inverse of H(Rao & Mitra, 1971):

β=H{Y:ð3Þ

2.2 Ensemble Techniques

2.2.1 Multiple Classiﬁers Technique

The multiple classiﬁer technique relies on the simple idea that the combination of

multiple classiﬁers leads to higher classiﬁcation prediction and efﬁciency than the

single classiﬁer. This approach is equivalent to the wisdom of crowds: the combined

opinion of diverse and independent experts usually outperforms the opinion of single

individuals. According to Kitter et al. (1998), the multiple classiﬁer technique

achieves higher efﬁciency when learners generalize in different ways, i.e., the

diversity of the ensemble is generated. As ELM is based on the random initialization

of internal bias and weights, each learner will be different; there is diversity in the

ensemble. Therefore, the forecast of several ELMs will be combined using majority

voting to produce the ﬁnal decision rule. Figure 1shows the general architecture of

the multiple classiﬁer.



Predicting Corporate Failure Using Ensemble Extreme Learning Machine 111

The classiﬁers C

(X),...,C

(X) are built based on the data set {(x

), (x

...,(x

)}. Each classiﬁer provides an output b

yMthat will be combined into the

ﬁnal output b

2.2.2 Bagging

Bagging (short for bootstrap aggregating) is one of the primal ensemble techniques

(Breiman, 1996). Its popularity lies in the fact that it is intuitive and simple to

implement, with notably good performance. Bagging generates the essential diver-

sity to create the ensemble process that manipulates the training set. In this regard,

the training set samples are randomly resampled in order to generate several different

bags of samples. Thus, each bag represents a set of training samples. Finally, the

base classiﬁer is applied to each bag, and the output classiﬁcation is made by a

majority vote of all the base classiﬁer results.

Bagging technique generates an improvement in generalization performance due

to the reduction in variance while maintaining steady or only slightly increasing the

bias, in particular, when it is applied to weak classiﬁers (Grandvalet, 2004). The

bagging algorithm can be expressed as follows:

Given a data set {(x

), (x

), ...,(x

)} .

1. Repeat for i=1, 2, ...,I.

(a) Build a bootstrap sample x

1,y

1,x

2,y

2,...,x

n,y

nby randomly

selecting ntimes with replacement from the data {(x

), (x

), ...,(x

)}.

(b) Fitting the bootstrapped classiﬁer C

on corresponding bootstrap sample.

2. Calculate the output of the ﬁnal classiﬁer:

CxðÞ=I-1I

iCixðÞ:ð4Þ

2.2.3 Boosting

Unlike the bagging technique, the boosting technique combines inaccurate and

relatively weak rules to produce highly accurate predictions. That is, it progressively

gives more weight to observations that have been misclassiﬁed by previously

generated classiﬁers in order to generate new classiﬁers and then combines the

classiﬁers of different iterations with weighted voting to make ﬁnal predictions.

Since numerous algorithms for boosting have been proposed, we use the Adaboost

algorithm (Freund & Schapire, 1996) which is one of the most popular boosting

techniques applied to pattern recognition (Verikas et al., 2010). The Adaboost

algorithm can be described as follows:







112 D. Veganzones

Given a data set {(x

), (x

), ...,(x

)} .

1. Initialize the weight vector of the training set:

W1iðÞ=1

Nfor i=1, ...,N:ð5Þ

2. For t=1, ...,T,

(a) Train the weak classiﬁer C

on the weighted training samples.

(b) Calculate the sum of weighted errors of C

εt=N

i=1Wt

i,Yi≠CtXi

ðÞ:ð6Þ

αt=1

2ln 1-εt

εt

:ð7Þ

(d) Update the weights:

Wtþ1

i=Wt

iexp -αtYiCtXi

ðÞðÞ

,ð8Þ

where Z

is a normalization factor.

3. Output:

fxðÞ=sign T

t=1αtCtxðÞ :ð9Þ

2.2.4 Random Subspace

The random subspace (Ho, 1998) bases its ensemble process on the modiﬁcation of

the feature space. That is, it creates different bags of training samples by randomly

selecting features drawn for the initial feature set that characterizes each sample. The

training sample X

(i=1, ...,n) in the training set X=(X

,...,X

)isap-dimen-

sional vector X

=(x

,...,x

), where prepresents the feature components.

Within the random subspace, the k-dimensional subspace is randomly selected

from the original p-dimensional feature space, k<p. The new learning samples

Xb=Xb

1,Xb

2,...,Xb



in a k-dimensional subspace Xb

i=xb

i1,xb

i2,...,xb

,

Prev xðÞ=

y2-1; 1

b=1δsgn CbxðÞ

ðÞ

,y:ð10Þ

where xb

ij j=1, ...,rðÞ,are built and then, the classiﬁers in the random subspace X

are combined using majority voting to create the ﬁnal decision rule. Thus, the

random subspace can be organized as follows:

Predicting Corporate Failure Using Ensemble Extreme Learning Machine 113

1. Repeat btimes, with b=1, 2, ...,B

(a) Randomly select a k-dimensional subspace X

among the initial p-dimen-

sional feature space X.

(b) Design a classiﬁer C

(x) using the sample X

2. Combine the forecast of C

(x) classiﬁers using majority voting to a ﬁnal

decision rule.

argmax XB

3 Experimental Design

3.1 Data

Our empirical study uses non-listed French ﬁrms taken from the Diane database

created by Bureau Van Dijk. The French companies must submit annual reports to

the French Commercial Court under French law provide accounting and income

statements to the Bureau Van Dijk authority. We drew ﬁrms from all sectors of

activity (excluding ﬁnancial companies) for the years 2016–2018, allowing us to

examine the model’s capacity to create good prediction rules in a real-world scenario.

The Diane database provides the information on whether ﬁrms have failed or remain

healthy; in the case of failure, it also provides the date. A ﬁrm is considered to be failed

if it proceeded to be liquidated or reorganized, and non-failed ﬁrms were those that

continued their activity for at least a year after the period studied. We decided to be

conservative in the selection of non-failed ﬁrm in order to avoid the inclusion of healthy

companies that may suddenly fail and ensure a reliable sample that does not fail.

Moreover, ﬁrms that presented missing values in their ﬁnancial statement, as well as

outliers, were excluded to ensure the prediction model stability. Consequently, the

collected dataset is composed of 3000 failed and 3000 non-failed ﬁrms.

Corporate failure is a rare phenomenon in the real world, so failed ﬁrms are clearly outnumbered

by non-failed ones. That is why the sample selection process becomes a signiﬁcant paradigm. If one

design a model based on the actual population, the dataset must be imbalanced. However, this

procedure has a main drawback: it is likely to lead to signiﬁcant degradation of the prediction

performance due to low percentage of failed ﬁrm in the entire sample (López et al., 2013;Shajalal

et al., 2021). Therefore, we collect a stratiﬁed sample with same observations of failed and

non-failed based on matched pair technique (Ciampi, 2015), in which failed ﬁrms are matched

with non-failed ﬁrms according to industry sector, size, and ﬁrm age.

114 D. Veganzones

To minimize the bias effect and sample variability that might inﬂuence the model

prediction performance, we carried out a tenfold cross-validation method in which

the dataset is split into ten distinct training and test set in order to learn and evaluate

the model prediction. This procedure was repeated ten times to ensure the reliability

of our results. Therefore, the ﬁnal prediction performance is calculated as the

average of 100 testing results.

3.2 Variables

Financial dimensions characterize the main explanatory factors for corporate failure.

Therefore, the balance sheets and income statements of the collected ﬁrms were used

to calculate 30 ﬁnancial ratios to use as explanatory variables. This representation

layer is important because it guarantees that the variables, we have used actually

represent all aspects of the phenomenon.

The initial set of ﬁnancial ratios that we compute includes at least four indicators

representing six categories: liquidity, solvency, proﬁtability, ﬁnancial structure,

turnover, and activity. These variables are presented in Table 1.

However, using all ﬁnancial ratios may result in very high-dimensional feature

space, which may reduce model predictive capability. Therefore, a variable selection

process has been performed in order to choose a subset of the most relevant ﬁnancial

ratios. Following the study by Kainulainen et al. (2011), a feed-forward variable

selection process was performed to retain the necessary information for prediction.

3.3 Evaluation Metrics

The evaluation criteria of our experiments are adopted from standard measures

established in the ﬁeld of prediction (Shahriare et al., 2021). These measures include

average accuracy, type error I, and type error II. The formula of these measures

provided below can be explained with respect to the confusion matrix shown in

Table 2.

Accuracy =TP þTN

TP þFP þFN þTN ,ð11Þ

Type -I error =TP

TP þFN ,ð12Þ

Type -II error =TN

TN þFP :ð13Þ

In addition to these evaluation metrics, we also used the area under the receiver

operating characteristic curve (AUC) to estimate the model performance. This is a

graphical plot used to represent the model performance while changing the cutoff

value. In this case, the proportion of true positive and false positive are plotted on the

x-axis and y-axis of the curve. AUC has become a widely used evaluation metric in

corporate failure prediction because it is insensitive to the matrix of misclassiﬁcation

cost

to assess the discrimination ability of a model. In summary, two classiﬁers can

be easily compared according to differences in the ROC curve performance. A

classiﬁer should get as close to the top left corner as possible, where its value will

be close to 1.

Predicting Corporate Failure Using Ensemble Extreme Learning Machine 115

Table 1 Initial set of variables

Proﬁtability Liquidity

X1 Proﬁt before Tax/Shareholders’Funds X16 Cash/total assets

X2 Net income/shareholders’funds X17 Current assets/current liabilities

X3 EBITDA/Total assets X18 Current assets/total debts

X4 EBIT/Total assets X19 Quick assets/Total assets

X5 Net income/Total assets X20 (Cash +Marketable securities)/Total sales

Financial structure Turnover

X6 Shareholder’s funds/Total assets X21 Inventory/Total sales

X7 Total debt/shareholders’funds X22 Net operating working /Total sales

X8 Total debt/Total assets X23 Accounts receivable/Total sales

X9 Net operating working/Total assets X24 Accounts payable/Total sales

X10 Long term debt/Total assets X25 Current assets/Total sales

Solvency Activity

X11 Financial expenses/Total sales X26 Cash ﬂow/total sales

X12 Labor expenses/Total sales X27 Total sales/total assets

X13 Financial debts/equity X28 Value added/total sales

X14 Financial expenses/EBITDA X29 Net income/value added

X15 Financial expenses/net income X30 EBITDA/Total sales

EBIT, earnings before interest and taxes; EBITDA, earning before interest, taxes, depreciation, and

amortization

Table 2 Confusion matrix for the prediction of corporate failure

Actually

Failed Healthy

Prediction Failed True positive (TP) False positive (FP)

Healthy False negative (FN) True negative (TN)

With the data set mentioned above, a cross-validation loop (tenfold cross-

validation repeated ten times) was performed to estimate the average evaluation

measures. To compare the classiﬁer performance, Demšar (2006) recommends a

The misclassiﬁcation of a failed ﬁrm (predict that a ﬁrm is healthy when it fails) represent a loss in

capital, while the misclassiﬁcation of a healthy ﬁrm (predict that a ﬁrm is failed when it survives)

represents only a loss of commercial bargain. That is why, misclassiﬁed a failed ﬁrm is considered

to be more costly.

Wilcoxon signed ranks non-parametric test because it only assumes limited com-

mensurability and can be applied to prediction accuracy, misclassiﬁcation errors or

any other evaluation metric. It is expressed as follows:

116 D. Veganzones

Given R

be the sum of ranks when the second classiﬁer outperforms the ﬁrst one,

be the sum of ranks for the opposite and the ranks of d

=0 are split evenly

among the sums:

Rþ=X

di>0

rank di

ðÞþ

di=0

rank di

ðÞ,ð14Þ

R-=

di<0

rank di

ðÞþ

2di=0

rank di

ðÞ:ð15Þ

Let Tbethe smaller of the sums, T=min (R

), the normal approximation can

be used and the following statistic is used to calculate the z-statistics with a

corresponding p-value:

z=T-nnþ1ðÞ

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

nnþ1ðÞ2nþ1ðÞ

q:ð16Þ

However, Garcia and Herrera (2008) caution that several repeated pairwise

comparison tests between algorithms conducted by us may lead to loss of control

over family-wise errors.

4 Results

Experimental analysis is designed to compare the prediction ability of different

ensemble methods based on extreme learning machine classiﬁer. Table 3indicates

the evaluation metrics achieved to assess the performance of the methods. Further-

more, this table is complemented by Table 4, which highlights whether the differ-

ences between the methods are statistically signiﬁcant.

We ﬁrst analyze the overall performance of the methods. Boosting ELM and

Bagging ELM achieve the best mean accuracy values, 82.2% and 82.6%, respec-

tively, while Random subspace ELM attains mean accuracy value of 81.7% and that

of 81.4% is achieved with Multiple ELM. All ensemble methods are more accurate

than the single ELM (80.4% of the mean accuracy). Thus, it conﬁrms that ensemble

ELM methods produce greater predictive power compared to a single ELM

Appendix 1 shows the results on the database using ELM and ELM-ensemble methods. Figures 2

and 3indicates the testing results with different number of hidden nodes and the average classiﬁ-

cation error of the ELM-ensemble methods as a function of the number of ensemble members.

classiﬁcation. The fact that Bagging and Boosting ensembles lead to the best

reduction in the generalization error is not entirely surprising, as it is well

documented their robustness to overﬁtting (Xiao et al., 2013; González et al.,

2020). In contrast, variation of the parameters of the classiﬁers, such as Multiple

ensemble and Random Subspace, can generate greater diversity (Bi, 2012). None-

theless, the information perceived by the varying diversity does not generate con-

sistent guidance so that the ensemble classiﬁer can obtain a good generalization. On

the whole, the key of Boosting and Bagging is that they build a set of diverse

classiﬁers, while they beneﬁt from the balance between diversity and accuracy,

which is an important determinant of the performance of ensemble classiﬁers.

Predicting Corporate Failure Using Ensemble Extreme Learning Machine 117

Table 3 Performance of different ELM-based ensemble methods

Accuracy Type-I error Type-II error AUC

ELM 80.4% 21.7% 17.5% 0.821

Multiple ELM 81.4% 20.3% 16.7% 0.834

Bagging ELM 82.6% 18.2% 16.5% 0.849

Boosting ELM 82.2% 18.8% 16.8% 0.842

Random subspace ELM 81.7% 20.0% 16.6% 0.836

Table 4 Signiﬁcance levels of a test of differences by method and evaluation metric

Multiple ELM Bagging ELM Boosting ELM Random subspace ELM

Accuracy

ELM 0.0866* 0.0001*** 0.0012*** 0.0338**

Multiple ELM 0.0463** 0.0971* 0.3372

Bagging ELM 0.2908 0.985*

Boosting ELM 0.2883

Type-I error

ELM 0.0976* 0.0001*** 0.0001*** 0.0652*

Multiple ELM 0.0179** 0.0751* 0.7871

Bagging ELM 0.5584 0.0386**

Boosting ELM 0.182

Type-II error

ELM 0.4275 0.0987* 0.4752 0.1255

Multiple ELM 0.7213 0.6531 0.6466

Bagging ELM 0.7889 0.6777

Boosting ELM 0.5133

AUC

ELM 0.0610* 0.0001*** 0.0001*** 0.0462**

Multiple ELM 0.0133** 0.1170 0.8674

Bagging ELM 0.2891 0.0811*

Boosting ELM 0.3746

*Signiﬁcant at 10% threshold; **Signiﬁcant at 5% threshold; ***Signiﬁcant at 1% threshold

Secondly, we ﬁnd no uniform improvement among the ensemble methods. If the

misclassiﬁcation errors are analyzed, Boosting ELM and Bagging ELM, here as

well, lead to lower misclassiﬁcation error for failed ﬁrms, 18.8% and 18.2%,

respectively, signiﬁcant at 1% threshold in comparison with ELM. In contrast, we

do not observe any signiﬁcant differences in misclassiﬁcation error for non-failed

ﬁrms across ensemble methods; rather, the mean type-II error ranges from 16.5%

with Bagging ELM and Random Subspace ELM to 18.8% with Bagging ELM.

118 D. Veganzones

Finally, the Bagging and Boosting ELM-based methods lead to higher AUC

values than the other ensemble methods, which is in line with the previous results. In

particular, Bagging ELM seems to be the most optimal ensemble method for

corporate failure prediction as results are signiﬁcantly better than those achieved

with the other ensemble methods, but with respect to Boosting ELM.

In sum, the better overall prediction of Bagging and Boosting methods over the

other ensemble methods, as previously observed, is due to their capacity to better

identify failed ﬁrms. The superiority of Bagging ELM is based on the creation of a

unique training set for each ensemble member because the perturbation generated in

the learning set causes a signiﬁcant change in the prediction constructed. As a

model’s prediction is order-correct for most of the replicated observation, the

bagging-based ELM can be transformed into a nearly optimal predictor, in particu-

lar, for failed ﬁrms. Furthermore, one of major reasons why boosted ELM better

identiﬁes failed ﬁrms may be due to the fact that the new classiﬁer generation gives

more relevance to misclassiﬁed observation, mostly failed ﬁrms. That is, the likeli-

hood of instances that have been misclassiﬁed by the previously generated classiﬁer

increases, and the set of classiﬁers grows progressively diverse. This trend explains

why this method provides higher accuracy for the minority class without jeopardiz-

ing the accuracy of the majority class.

4.1 Further Validation

In order to further evaluate the effectiveness of the ensemble extreme learning

machine for the corporate failure prediction task, a new data set has been collected.

In general, there is no universal accepted deﬁnition of corporate failure; bankruptcy,

the more severe form of failure, is commonly used. The popularity of bankruptcy as

the deﬁnition of failure is based on two concepts: on the one hand, it provides an

objective criterion to distinguish failed and non-failed ﬁrms, and, on the other hand,

the moment of failure can be dated when a ﬁrm ﬁlls in the bankruptcy procedure.

Therefore, the bankruptcy notion offers a discrimination criterion for obtaining a

well-deﬁned dichotomy, or at least, a representation of corporate failure, that can be

applied methodologically. Nonetheless, numerous studies (Sun et al., 2014;Brédart

et al., 2021) consider that corporate failure begins when a ﬁrm experiences ﬁnancial

distress. That is, when a ﬁrm encounters ﬁnancial difﬁculties or struggles to fulﬁll its

obligations. Accordingly, we collected a data set considering ﬁnancial distress as the

deﬁnition of corporate failure. We consider the criterion provided by Balcaen et al.

(2011), who deﬁne ﬁnancial distress as a ﬁrm with negative recurring proﬁt after

taxes over two consecutive years. Consequently, the collected dataset is composed

of 2500 failed and 2500 non-failed ﬁrms.

Predicting Corporate Failure Using Ensemble Extreme Learning Machine 119

Table 5 Performance of different prediction methods

Accuracy Type-I error Type-II error AUC

ELM 78.2% 24.7% 18.9% 0.790

Multiple ELM 79.5% 23.0% 18.0% 0.804

Bagging ELM 81.1% 20.7% 17.1% 0.824

Boosting ELM 80.5% 21.4% 17.6% 0.812

Random subspace ELM 80.0% 22.1% 17.9% 0.808

Table 6 Signiﬁcance levels of a test of differences by method and evaluation metric

Accuracy

Multiple ELM Bagging ELM Boosting ELM Random subspace ELM

ELM 0.0753* 0.0001*** 0.0032** 0.0217**

Multiple ELM 0.0265** 0.1333 0.2766

Bagging ELM 0.1267 0.0836*

Boosting ELM 0.3045

Type-I error

Multiple ELM Bagging ELM Boosting ELM Random subspace ELM

ELM 0.0592* 0.0001*** 0.0001*** 0.0154**

Multiple ELM 0.0144** 0.0869* 0.1936

Bagging ELM 0.1709 0.0935*

Boosting ELM 0.2423

Type-II error

Multiple ELM Bagging ELM Boosting ELM Random subspace ELM

ELM 0.2611 0.0348** 0.0107 0.2414

Multiple ELM 0.2560 0.3987 0.5612

Bagging ELM 0.6214 0.3521

Boosting ELM 0.3951

AUC

Multiple ELM Bagging ELM Boosting ELM Random subspace ELM

ELM 0.0509* 0.0001*** 0.0028*** 0.0131**

Multiple ELM 0.0106** 0.1635 0.5145

Bagging ELM 0.0958* 0.0439**

Boosting ELM 0.3153

*Signiﬁcant at 10% threshold; **Signiﬁcant at 5% threshold; ***Signiﬁcant at 1% threshold

The results presented in Tables 5and 6are consistent with those of the previous

ones. Boosting ELM and Bagging ELM achieve the highest accuracy values, in

particular, due to their effectiveness in the reducing the type-I error in comparison to

To design the prediction methods, the same procedure used in Sect. 3.2 was followed. Then, they

were evaluated based on a 10-cross validation and using the abovementioned evaluation metrics.

the single ELM.

Moreover, it is important to mention that the prediction perfor-

mance of the methods in this data set is inferior to the previous one. Thus, it is more

arduous to differentiate failed ﬁrms from healthy ones in the initial steps of failure,

when ﬁrms just experience ﬁnancial distress. The literature documented that ﬁrms

have shown a certain resilience for a long time, even though their ﬁnancial situation

resembles to a bankrupt one (Iftikhar et al., 2021). In contrast, ﬁrms that seem

completely sound may suddenly fail. Therefore, the inability to know whether the

echoes of ﬁnancial distress may result in corporate failure makes it difﬁcult to

capture distinguishable factors that might reinforce model accuracy. That is why

the performance of models is lower when corporate failure is represented as ﬁnancial

distress than when it is deﬁned as bankruptcy.

120 D. Veganzones

5 Conclusion

In this study, we propose to evaluate several ensemble methods applied to corporate

failure prediction in order to improve the classiﬁcation performance of ELM. An

ensemble strategy that combines the predictions of individual models is more

performance-based than relying on the prediction capacity of a single model. Our

results conﬁrm that the Extreme Learning Machine-based ensemble is more accurate

and robust than the “individual best”ELM model using two real ﬁnancial datasets. In

particular, the ensemble methods used in this study increase, on average, the

classiﬁcation accuracy estimated for the single ELM by 1.6 and 2.1 percentage

points for the bankruptcy data and ﬁnancial distress data, respectively. An increase

in prediction performance of these magnitudes may seem modest, but the readers

need to understand that ﬁnancial institutions and banks can save a huge amount of

the limited ﬁnancial resources with decision technology that can increase the pre-

diction power by 2%.

As Bagging ELM and Boosting ELM give similar results –there is some

evidence that the bagging strategy is more effective for the prediction of corporate

failure using ELM –it is arduous to make a design recommendation for which

method is more optimal. However, we do notice that both methods, which operate by

taking a base learner and invoking it multiple times using different training sets, are

most effective in the ensemble ELM prediction method. We also notice that bagged

ELM is more computationally efﬁcient, as it requires 40–50 ensemble members,

while 60–70 members as necessary for the boosting ensemble.

Acknowledgments We sincerely thank Prof. Abedin and Prof. Hajek for their assistance.

The Appendix 2 shows graphically the testing results with different hidden nodes (Fig. 4) and the

average classiﬁcation error of ELM-ensemble methods as a function of ensemble members (Fig. 5).

Predicting Corporate Failure Using Ensemble Extreme Learning Machine 121

Appendices

Appendix 1

Fig. 2 Testing results for different hidden nodes in ELM for bankruptcy data

Fig. 3 Average classiﬁcation errors of the Ensemble ELM methods by ensemble members for

bankruptcy data

122 D. Veganzones

Appendix 2

Fig. 4 Testing results for different hidden nodes in ELM for ﬁnancial distress data

Fig. 5 Average classiﬁcation errors of the Ensemble ELM methods by ensemble members for

ﬁnancial distress data

Predicting Corporate Failure Using Ensemble Extreme Learning Machine 123

References

Abedin, M. Z., Chi, G., Colombage, S., & Moula, F. E. (2018). Credit default prediction by using a

support vector machine and a probabilistic neural network. Journal of Credit Risk, 14(2), 1–27.

Abedin, M. Z., Hassan, M. K., Petr, H., & Uddin, M. M. (2021). Machine learning in ﬁnance and

accounting. In The essentials of machine learning in ﬁnance and accounting, Taylor & Francis.

Abedin, M. Z., Chi, G., Hajek, P., & Tong, Z. (2022). Combining weighted SMOTE with ensemble

learning for the class-imbalanced prediction of small business credit risk. Complex & Intelligent

Systems.https://doi.org/10.1007/s40747-021-00614-4

Akusok, A., Veganzones, D., Miche, Y., Björk, K. M., Du Jardin, P., Severin, E., & Lendasse,

A. (2015). MD-ELM: Originally mislabeled samples detection using OP-ELM model.

Neurocomputing, 159, 242–250.

Altman, E. I. (1968). Financial ratios, discriminant analysis and the prediction of corporate

bankruptcy. The Journal of Finance, 23(4), 589–609.

Balcaen, S., Manigart, S., & Ooghe, H. (2011). From distress to exit: Determinants of the time to

exit. Journal of Evolutionary Economics, 21, 407–446.

Beaver, W. H. (1966). Financial ratios as predictors of failure. Journal of Accounting Research, 4,

71–111.

Bi, Y. (2012). The impact of diversity on the accuracy of evidential classiﬁer ensembles. Interna-

tional Journal of Approximate Reasoning, 53(4), 584–607.

Brédart, X., Séverin, E., & Veganzones, D. (2021). Human resources and corporate failure

prediction modeling: Evidence from Belgium. Journal of Forecasting, 40(7), 1325–1341.

Breiman, L. (1996). Bagging predictors. Machine Learning, 24(2), 123–140.

Ciampi, F. (2015). Corporate governance characteristics and default prediction modeling for small

enterprises: An empirical analysis of Italian ﬁrms. Journal of Business Research, 68(5),

1012–1025.

Demšar, J. (2006). Statistical comparisons of classiﬁers over multiple data sets. The Journal of

Machine Learning Research, 7,1–30.

Freund, Y., & Schapire, R. E. (1996). Experiments with a new boosting algorithm. In Thirteenth

International Conference on Machine Learning (pp. 148–156). IEEE.

Garcia, S., & Herrera, F. (2008). An extension on “statistical comparisons of classiﬁers over

multiple data sets”for all pairwise comparisons. Journal of Machine Learning Research,

9(12), 2677–2694.

González, S., García, S., Del Ser, J., Rokach, L., & Herrera, F. (2020). A practical tutorial on

bagging and boosting based ensembles for machine learning: Algorithms, software tools,

performance study, practical perspectives and opportunities. Information Fusion, 64, 205–237.

Grandvalet, Y. (2004). Bagging equalizes inﬂuences. Machine Learning, 55(3), 251–270.

Ho, T. K. (1998). The random subspace method for constructing decision forests. IEEE Trans-

actions on Pattern Analysis and Machine Intelligence, 20(8), 832–844.

Huang, G. B., Zhu, Q. Y., & Siew, C. K. (2006a). Extreme learning machine: Theory and

applications. Neurocomputing, 70(1), 489–501.

Huang, G. B., Chen, L., & Siew, C. K. (2006b). Universal approximation using incremental

constructive feedforward networks with random hidden nodes. IEEE Transactions on Neural

Networks, 17(4), 879–892.

Iftikhar, A., Purvis, L., & Giannoccaro, I. (2021). A meta-analytical review of antecedents and

outcomes of ﬁrm resilience. Journal of Business Research, 135, 408–425.

Kainulainen, L., Miche, Y., Eirola, E., Yu, Q., Frénay, B., Séverin, E., & Lendasse, A. (2011).

Ensembles of local linear models for bankruptcy analysis and prediction. Case Studies in

Business, Industry and Government Statistics, 4(2), 116–133.

Kim, M. J., & Kang, D. K. (2010). Ensemble with neural networks for bankruptcy prediction.

Expert Systems with Applications, 37(4), 3373–3379.

Kittler, J., Hatef, M., Duin, R. P. W., & Matas, J. (1998). On combining classiﬁers. IEEE Trans-

actions on Pattern Analysis and Machine Intelligence, 20(3), 226–239.

124 D. Veganzones

Kumar, P. R., & Ravi, V. (2007). Bankruptcy prediction in banks and ﬁrms via statistical and

intelligent techniques: A review. European Journal of Operational Research, 180(1), 1–28.

López, V., Fernández, A., García, S., Palade, V., & Herrera, F. (2013). An insight into classiﬁcation

with imbalanced data: Empirical results and current trends on using data intrinsic characteristics.

Information Sciences, 250, 113–141.

Moula, F. E., Chi, G., & Abedin, M. Z. (2017). Credit default prediction modeling: An application

of support vector machine. Risk Management, 19(2), 158–187.

Ouenniche, J., & Tone, K. (2017). An out-of-sample evaluation framework for DEA with applica-

tion in bankruptcy prediction. Annals of Operations Research, 254(1), 235–250.

Rao, C. R., & Mitra, S. S. K. (1971). Generalized inverse of matrix and its application (Wiley

Series in Probability and Mathematical Studies). Wiley.

Rokach, L. (2010). Ensemble-based classiﬁers. Artiﬁcial Intelligence Review, 33(1), 1–39.

Séverin, E., & Veganzones, D. (2021). Can earnings management information improve bankruptcy

prediction models? Annals of Operations Research, 306(1), 247–272.

Shahriare S, Khair A, Abedin MZ (2021, December 19–21). Performance analysis of machine

learning techniques that predict hotel booking cancellations in hospitality industry. In ICCIT

2020: 23rd International Conference on Computer and Information Technology, Dhaka.

Shajalal, M., Abedin, M. Z., & Uddin, M. M. (2021). Handling class imbalance data in business

domain. In: The essentials of machine learning in ﬁnance and accounting. Taylor & Francis.

Sun, J., Li, H., Huang, Q. H., & He, K. Y. (2014). Predicting ﬁnancial distress and corporate failure:

A review from the state-of-the-art deﬁnitions, modeling, sampling, and featuring approaches.

Knowledge-Based Systems, 57,41–56.

Veganzones, D., & Severin, E. (2020). Corporate failure prediction models in the twenty-ﬁrst

century: A review. European Business Review, 33(2), 204–226.

Verikas, A., Kalsyte, Z., Bacauskiene, M., & Gelzinis, A. (2010). Hybrid and ensemble-based soft

computing techniques in bankruptcy prediction: A survey. Soft Computing, 14(9), 995–1010.

Xiao, T., Zhu, J., & Liu, T. (2013). Bagging and boosting statistical machine translation systems.

Artiﬁcial Intelligence, 195, 496–527.

Yu, Q., Miche, Y., Séverin, E., & Lendasse, A. (2014). Bankruptcy prediction using extreme

learning machine and ﬁnancial expertise. Neurocomputing, 128, 296–302.

125

Assessing and Predicting Small Enterprises’

Credit Ratings: A Multicriteria Approach

Baofeng Shi

Abstract Credit ratings play a key role in helping ﬁnancial institutions to make loan

decisions and to reduce the ﬁnancial constraints on small and medium-sized enter-

prises. However, small enterprises have made it difﬁcult for ﬁnancial institutions

such as commercial banks to accurately determine their credit risk, creating salient

loan difﬁculties, due to the short duration, high frequency, urgent demand for credit,

and small amount of their loans. In order to alleviate the difﬁculties of ﬁnancing

small businesses, this paper develops a new approach for the assessment of credit

risk in small enterprises by combining high-dimensional attribute reduction methods

with fuzzy decision-making methods. Based on 687 small enterprises in a regional

commercial bank of China, we ﬁnd 17 indicators that have a signiﬁcant impact on

the default risk of small enterprises. Then, it utilizes TOPSIS together with fuzzy

C-means to grade the credit ratings of enterprises requesting loans. The standard

discrimination and ROC curve dual tests resulted in the prediction accuracy of the

standard indicator system reaching 85.40 percent and 90.09 percent, respectively,

indicating the strong default discrimination of this rating system and its practicability

in commercial banks and other ﬁnancial institutions.

Keywords Credit rating · Default risk · Fuzzy C-means · Small enterprises

1 Introduction

China is the world’s largest developing country, and small and medium-sized

enterprises have developed rapidly. According to statistics, in 2021, Chinese

SMEs contribute more than 80% of national employment, 60% of gross domestic

B. Shi (✉)

College of Economics and Management, Northwest A&F University, Xianyang, Shaanxi, China

Research Center on Credit and Big Data Analytics, Northwest A&F University, Xianyang,

Shaanxi, China

e-mail: shibaofeng@nwsuaf.edu.cn

©The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

M. Z. Abedin, P. Hajek (eds.), Novel Financial Applications of Machine Learning

and Deep Learning, International Series in Operations Research & Management

Science 336, https://doi.org/10.1007/978-3-031-18552-6_8

product (GDP) (iResearch, 2021). Yet, small and medium-sized enterprises gener-

ally struggle to obtain ﬁnancing, especially loans, severely restricting their develop-

ment, due to unreliable ﬁnancial information, loans of enormous volume but for low

amounts, and diverse risks (Lu et al., 2022; Abedin et al., 2021; Ciampi & Gordini,

2013; Shi et al., 2016; Chi & Zhang, 2017; Ruan et al., 2018; Sun et al., 2022). To

alleviate these ﬁnancial difﬁculties, the Chinese Banking Regulatory Commission

and other agencies requested the establishment of an “Inclusive Finance Business

Division,”provide ﬁnancial services to small and micro businesses, and address

issues affecting agriculture, rural areas, and farmers, and strengthen credit risk

identiﬁcation, surveillance, early warning and assessment of borrowers (CBRC,

2015; SCPRC, 2016,2017).

126 B. Shi

Many scholars have conducted useful studies on the best way to assess the credit

risk of loan-granting enterprises, in terms of the establishment of credit scoring,

credit rating, and other systems. Dimensionless processing of statistics is typically

necessary before a rating system can be implemented (Shi et al., 2015). In reality, the

quantiﬁable ﬁnancial data of small enterprises are less and more text-based

non-ﬁnancial data. As a result, researchers often use subjective Delphi method or

analytic hierarchy process (AHPs) to process data without dimensions (Liang, 2007;

Shi et al., 2018).

Regarding the development of indicator systems, Altman constructed Z-score and

ZETA models based on ﬁnancial indicators such as return on assets and pretax

margins of asset interest to assess the probability of lender default (Altman, 1968;

Altman et al., 1977). Gu et al. (2017) combined (AHP) with data envelopment

analysis (DEA), using indicators such as the cash ratio, inventory turnover, and

accounts receivable turnover ratio from the perspective of ﬁnancial status, enterprise

development, credit status, and internet ﬁnancial status to predict defaults by enter-

prises that take out loans. This research has great reference value for creating a credit

rating indicator system for small enterprises, but little of it studies wholesale and

retail enterprises and uses distinctive default variables to forecast the credit outlook

of loan customers.

Credit scoring models can be constructed using three methods: metrological

statistics, fuzzy systems, and artiﬁcial intelligence. Metrological statistics consist

of discriminant analysis, logistic regression, and linear regression (Reichert et al.,

1983; Yurdakul & Iç, 2015; Iç & Yurdakul, 2010). Artiﬁcial intelligence methods

include artiﬁcial neural nets (Marcano-Cedeño et al., 2011; Rui & Mendes, 2017;

Chi et al., 2017), support vector machine (Hens & Tiwari, 2012; Harris, 2015;

Tomczak & Zięba, 2015; Abedin et al., 2018;2019a,b), a decision tree (Zhu &

Hu, 2013; Florez-Lopez & Ramon-Jeronimo, 2015; Bahnsen et al., 2015; Zhang

et al., 2017; Chern et al., 2021), ensemble learning (Abedin et al., 2022), and so

forth. Recently, some academics have combined these methods with fuzzy evalua-

tions and subsequently developed credit rating systems. Akkoç (2012) combined

fuzzy evaluation and artiﬁcial intelligence to develop a credit rating system using a

hybrid adaptive neuron fuzzy inference system predicting the risk of default of credit

card holders in Turkey. The empirical research shows that this model is better at

correctly averaged classiﬁcation and wrongly estimated classiﬁcation cost than liner

discriminant analysis, logistic regression, and artiﬁcial neural nets. Bai et al. (2019)

calculate the risk of default for farm lenders in a hybrid model using fuzzy C-means

(FCM) and fuzzy rough sets. This study reveals the determinants of loan defaults,

without grading their credit or including any decision function in their evaluation

results.

Assessing and Predicting Small Enterprises’Credit Ratings:... 127

To address this problem, some scholars have begun to divide consideration of

credit ratings of loan customers into three credit rating models: scoring intervals of

customer credit, establishing the threshold of default probability, and the loss given

default (LGD) of loan customers. The Industrial and Commercial Bank of China

(ICBC) (2005) divided the credit scores of its loan customers among 10 credit ratings

into AA, AA-, so forth. Florez-Lopez (2007) estimated the default probability

(PD) of loan applicants using statistical and artiﬁcial intelligence methods and

classiﬁed the applicants into ﬁve rating categories. Chi and Zhang (2017) employed

nonparametric models to construct a credit rating system speciﬁcally designed for

small enterprises. They evaluate the credit ratings of loan customers according to

their LGD. Therefore, credit rating models based on credit scoring intervals for

customers give different results than models based on the threshold of default

probability, so different loan approvers may give different results of credit rating

for loan customers with those credit scores. The reason is that scoring intervals and

the threshold of default probability are given ahead of time, and this increases the

subjectivity of the ratings. With regard to the credit rating method based on LGD, a

prerequisite is that the default loss of each customer must be known. However,

default loss data are not available for some small enterprises that have only recently

applied for loans, making this rating method infeasible.

Through our literature review, we ﬁnd that there is no existing research that has a

suitable rating indicator system to measure credit risk based on the loan character-

istics of small wholesale and retail enterprises. In fact, industry differences among

small enterprises lead to obvious heterogeneity in their estimation of loan and credit

risks. For example, the statistics on credit at commercial banks show that the average

maximum value of loans given to small enterprises in real estate development and

operations is as much as 17 million Yuan (about USD 2.50 million) and that of small

enterprises in wholesale and retail only amount to 0.41 million Yuan (Bank of

Dalian, 2014). When comparing these two types of companies in the same credit

risk system, even if the default model false positive is very low, the bank will suffer

completely different losses. Therefore, different credit rating models are required for

different industries, based on the fact that they are small enterprises, to distinguish

their credit risk from that of other kinds of enterprises.

In view of the foregoing, this paper makes three contributions to the literature.

First, in the category of credit rating, it adds to the literature by focusing on Chinese

small wholesale and retail enterprises. Second, by establishing suitable credit rating

models for small wholesale and retail enterprises, it offers a decision-making

reference for credit rating by commercial banks, microcredit organizations, and

these enterprises. Third, we propose a credit scoring measurement process by

using triangular fuzzy numbers for non-ﬁnancial data at small wholesale and retail

enterprises, which helps to avoid the subjectivity and randomness caused by exper-

tise scoring and makes the quantiﬁed processed qualitative indicator more accurate.

128 B. Shi

The paper is organized as follows. Section 2introduces credit rating models for

small enterprises. Section 3builds the rating system based on credit data for

687 small wholesale and retail enterprises seeking loans from an urban commercial

bank in China. Section 4offers our main conclusion and lists the innovative aspects

of this paper.

2 Methodology

First, we set up an assessment system based on the characteristics of small wholesale

and retail loans. Second, TOPSIS is used to obtain credit scores based on the

indicator weights computed as entropy weights. Finally, fuzzy C-means is used to

evaluate the credit ratings of loan customers. The framework can be seen in Fig. 1.

2.1 Establishment of a Credit Rating System

The establishment of this credit rating system is done in two steps. Firstly, initial data

must be standardized to eliminate incompatibility between different measurement

measures. Second, probit regression and partial correlation analysis are combined to

create quantitative screening to reduce the number of indicators.

Pre-Processing of Indicator Data

1. Pre-Processing of Qualitative Indicator

Qualitative indicators cannot be directly quantiﬁed but, rather, are described

narratively. For instance, the indicator for education background has ﬁve possible

values: “Primary school diploma,”“junior high school diploma,”“senior high

school diploma,”“junior college diploma,”and “bachelor’s degree or above.”

Qualitative indicators have an advantage similar to that of triangular fuzzy

numbers in how they process data with diverse characteristics. To quantify the

qualitative indicators, they must be transformed to triangular fuzzy numbers

according to their semantics; then, defuzziﬁcation is used, that is, triangular

fuzzy numbers are transformed to ﬁxed values.

Let Abe a fuzzy set for x2U,ifμ

(x)2[0, 1], then μ(x) is the membership of

xto U, and μ

represents the membership function of x. Further, le land ube the

lower and upper limit of the fuzzy number, respectively, and let mbe the median

value, then the fuzzy number (l,m,u) can be shown in Fig. 2. Its membership

function μ

is presented in Eq. (1) (Promentilla et al., 2008). Typically, three, ﬁve,

and seven triangular fuzzy numbers are used (Cheng et al., 2008; Khalili-

Damghani et al., 2013;Wang et al., 2016), as illustrated in Figs. 3,4,and5

(Chai et al., 2019).

Assessing and Predicting Small Enterprises’Credit Ratings:... 129

Part 2:

Calculate

customers

credit scoring

Part 3:

Divide

customers

credit ratings

Classify customer’s credit rating using FCM algorithm

Compute credit score of small wholesale and retail enterprise adopting

TOPSIS method

Calculate the indicator's weight by using entropy weighting method

Calculate the partial correlation coefficient of indicators

Step 1:

Indicators data

preprocessing

Establish the Probit regression equation of the default state Y and the

evaluation indicators X

Calculate LR value and Sig value of each regression

Calculate F value of

indicators and delete the

indicator which F value

is smallest

First round screening

based on the partial

correlation analysis:

Select indicator

which the F value is

larger in the two

indicators with

greater than 0.7,

ensuring that the

indicator system

wouldn't reflect the

duplicated

information

Step 2:

Partial

correlation

analysis

Significant probability

of each indicator

sig<0.01

Delete the indicator

which the sig value

is the largest

Second round

screening based on

the Probit

regression:

Ensure that the

selected indicators

effectively

differentiate default

and non-default

small wholesale

and retail

enterprises

Step 3 : Probit

regression

(quantitative

screening)

The credit rating indicator system of small wholesale and retail enterprises

Yes

Mass-selection indicator set

Standardized evaluation indicators (defuzzification)

Transforme qualitative indicators into quantitative indicators (Triangular

fuzzy numbers)

|<0.7

Part1:

Establish the

credit rating

indicator

system

Fig. 1 Framework of the credit rating model

130 B. Shi

Fig. 2 Triangular fuzzy

numbers (TFNs)

(x)

0lmu

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Fig. 3 TFNs with three classiﬁcations

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Fig. 4 TFNs with ﬁve classiﬁcations

Assessing and Predicting Small Enterprises’Credit Ratings:... 131

Calculation of a new

membership matrix

Yes

Initializing cluster

center c0,1

···

Initializing cluster

center c

0,2

Initializing cluster center

0,9

J< ε

Cluster center c

t,1

···

Cluster center c

t,9

Cluster center c

t,2

Enterprises

Initialization of

basic paramet ers ,

numbers of

clusters and

Membership

matrix

Iterations <

initial assumed

value T

Yes

Fig. 5 The framework for dividing credit ratings using the FCM method

μAxðÞ=

0x<l,

x-l

m-ll<x<m,

u-x

u-mm≤x≤u,

0x>u,

;

:ð1Þ

Let A

max

be the defuzziﬁed value, then when combined with Eq. (1), A

max

is given

as follows (Wu et al., 2016):

Amax =lþmþuðÞ=3:ð2Þ

2. Pre-Processing of Quantitative Indicator

Quantitative indicators usually include four types of indicators, namely

positive, negative, interval, and moderating indicators. We can use the

max-min standardization for the indicators (Chi & Zhang, 2017; Shi et al.,

2018; Abedin et al., 2019a,b); to avoid repetition, it is not described here.

Reduction of Attributes

1. The First Indicator Screening Based on Partial Correlation Analysis

In the same standard layer, partial correlation analysis (PCA) is used to remove

redundant indicators. Let x

be the value of indicator ifor enterprise j,r

be the

correlation coefﬁcient between indicators iand k, then r

is deﬁned as follows:

P

132 B. Shi

rik=

j=1xij -

xixij -

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

j=1xij -



q ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

j=1xij -



q,ð3Þ

where nis the number of enterprises, and 

xiis the average value of indicator i.

Suppose that Ris the correlation matrix composed of r

, and mis the number of

variables at the criterion level. The correlation matrix Ris given as follows:

r11 r12 ⋯r1m

r21 r22 ⋯r2m

⋮⋮⋯⋮

rm1rm2⋯r

5:ð4Þ

The inverse matrix Cof the correlation matrix Ris:

C=R-1=

c11 c12 ⋯c1m

c21 c22 ⋯c2m

⋮ ⋮⋯⋮

cm1cm2⋯c

5:ð5Þ

Then, the partial correlation coefﬁcient of indicator iand indicator kcan be

obtained:

ik =

-cik

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

ciickk

p:ð6Þ

The larger the partial correlation coefﬁcient r0

ik, the stronger the relativity between

indicators iand k. When r0

>0:7, Ftest (Nami & Shajari, 2018) is employed to

perform the evaluation of the two indicators. Subsequently, the indicator with a

lower Fvalue is removed.

2. The Second indicator Screening Based on Probit Regression

In the same standard layer, the maximum likelihood function is employed to

obtain the probit regression coefﬁcients between the mindicators and the default

, and to determine the LR statistics of each indicator. Using χ

, we remove the

indicator with the largest sig but that shows the least remarkable effects on

defaults among the indicator with a signiﬁcance probability (Sig >0.01), and

complete the screening of the ﬁrst indicator. The remaining m-1 indicators,

will be screened in the same manner as above until the corresponding signiﬁ-

cance probability of each indicator fails to exceed 0.01, i.e., Sig ≤0.01. Then the

indicator screening is done. Now, the remaining indicators can all signiﬁcantly

distinguish the defaults of small enterprises. The speciﬁc resolution equation is

as follows.

Z



Assessing and Predicting Small Enterprises’Credit Ratings:... 133

Let X

=(x

1j,

,...,x

) be the row vector of enterprise j;β=(β

,β

,...,β

)

be the regression coefﬁcient vector of indicators; mdenotes the number of

indicators; φ(z

) is the standardized normal cumulative distribution function,

P(Y

=1) indicates the probability of default; and z

=α+X

β. Then,

j=1



=ϕzj

=

-1

ﬃﬃﬃﬃﬃ

2π

pℓ

-s2

2ds:ð7Þ

The maximum likelihood method can be used to predict the indicators in the

probit model. Its log-likelihood function is deﬁned as follows:

max ln L =Xn

j=1yjln ϕzj

þ1-yj



ln 1 -ϕzj

:ð8Þ

In Eq. (8), the larger the log-likelihood function LnL, the more accurate estimate

of default Y

Suppose that LR

is the LR statistic value for indicator k,σ

βk

is the standard error

of regression coefﬁcient β

βkis the estimated parameter value, bσβkis the standard

error of the estimated parameter value, and b

βkas well as bσβkare independently the

estimated value and standard error beyond constraints. Then:

LRk=-2 log L~

βk,~σ2

βk



-log Lb

βk,bσ2

βk

h:ð9Þ

2.2 Solution to Credit Scoring

Entropy weight is a method of describing the differences in information between

indicators based on entropy in information in evaluated statistics; it has often been

used in evaluation of complex systems (Chi & Zhang, 2017; Bai & Zhao, 2022). In

this section, entropy is used to calculate the evaluation indicator weight W=(w

)in

the ﬁrst place; then TOPSIS is used to obtain credit scores (Yurdakul & Iç, 2015;Iç

& Yurdakul, 2010; Wang & Leng, 2021). The procedure is presented as follows:

Step 1: Obtain the best and worst scores of the indicators.

Suppose that bþ

iand b-

iare the best and worst scores of indicator i, respectively,

and b

is the score for enterprise j;so

bþ

i=max bij

ðÞ

,idenotes the ith positive indicator

min bij

ðÞ

,jdenotes the jth negative indicator

:ð10Þ



i=min bij

ðÞ

,idenotes the ith positive indicator

max bij

ðÞ

,jdenotes the jth negative indicator :ð11Þ



P P

134 B. Shi

Step 2: The standardized score is obtained, and the difference between the best

and worst scores are calculated. Suppose that dþ

j(and d-

j) are the differences

between the best (worst) score and the actual score of enterprise j. Then,

dþ

j=ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

i=1wibij -wibþ



q,d-

j=ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

i=1wibij -wib-

2

q:ð12Þ

Step 3: Independently solve for the difference between the best and worst scores

and the relative closeness of the credit scores. Suppose that c

is the relative closeness

of the score, and P

be the credit score:

Pj=cj=

jþdþ

:ð13Þ

Step 4: The credit score P

in Eq. (13) range from 0 to 1, which are not consistent

with the customary scoring regulations on a scale of 100. In view of this, we

standardize P

to render it in a period from 0 to 100.

Sj=Pj-min Pj



max Pj



-min Pj



×100, ð14Þ

where S

is the standardized credit score of enterprise j.

This paper employs default discrimination and a ROC curve to evaluate the

predictive performance of the system for small enterprises as follows: if the credit

score of a rating system meets the requirement that “all the credit scores of

non-defaulting small enterprises are higher than those of small defaulting enter-

prises”the stronger the evaluation ability of the indicator system on the defaults of

loan enterprises becomes, the fewer the losses of ﬁnancial institutions such as banks.

In agreement with Chi and Zhang (2017), the rationality of the indicator system is

determined.

c=1

mXm

j=1S1

j,ð15Þ

c=1

j=1S0

j,ð16Þ

sc=

j=1S1

jþ1

j=1S0

2,ð17Þ

where S0

cand S1

cdenote the average value of the credit scores of non-defaulting and

defaulting samples, respectively, S

=(S1

c+S0

c)/2.

ROC was ﬁrst applied by Sobehart and Keenan (2001) to evaluate the accuracy of

credit ratings. First, the sensitivity and speciﬁcity of the credit rating system are

obtained. Given that the number of correctly determined defaulting samples (y

=1)

ðÞ !

is TP (true positive); the number of incorrectly determined defaulting samples is FN

(false negative); the number of correctly classiﬁed non-defaulting samples (y

=0) is

TN (true negative); the number of incorrectly non-defaulting samples is FP (false

positive), sensitivity and speciﬁcity can be calculated as follows:

Assessing and Predicting Small Enterprises’Credit Ratings:... 135

Sensitivity =TP

TP þFN :ð18Þ

Specificity =TN

FP TN :ð19Þ

Then, sensitivity and speciﬁcity can be used to draw the ROC curve of the system.

The larger the area under the ROC curve, the stronger the system’s capacity to

recognize defaulting samples.

2.3 Dividing Credit Ratings of Loan Customers

In contrast to conventional cluster algorithms, fuzzy cluster algorithms do not

require strict identiﬁcation of objects belonging to speciﬁc classes, demonstrating

ﬂexible attribute requirements. Thus, it ﬁts the special requirement that the initial

indicator information is a value of a triangular fuzzy function. Therefore, this paper

follows Bai et al.’s(2019) fuzzy C-means (FCM) algorithm, in rating the credit of

small enterprises. The principle is shown in Fig. 5.

The FCM compares each sample with all clusters using real values u

, ranging

from 0 to 1, reﬂecting the degree of membership of indicator jin category i.

FCM divides the mvectors S

(j=1,2,,m) into cfuzzy clusters, and calculates

the center of each cluster so that non-similarity objective function is minimized. Its

objective function J(U,c

,⋯,c

) (Yu et al., 2010) is:

JU,c1,⋯,cc

ðÞ=Xc

i=1Xm

j=1uij



nd2xj,ci



,ð20Þ

where d(S

) is the Euclidean distance of the clustering center c

in the sample S

;

n2[1, 1) is the weighting indicator, controlling the shared degree of the classiﬁed

objects in the fuzzy category.

Its structure is shown as the following objective function



JU,c1,c2⋯,cc,λ1,⋯λm(Sun et al., 2022):

JU,c1,c2⋯,cc,λ1,⋯λm

ðÞ=JU,c1,c2⋯,cc

ðÞþ

j=1

λX

i=1

uij -1

i=1X

j=1

uij



nd2

ij þX

j=1

λX

i=1

uij -1

! ð21Þ

In this equation, λ

is the Lagrange multiplier; c

and u

are deﬁned as follows

(Demircan & Kahramanli, 2016):

P

136 B. Shi

ci=

j=1uij

nSj

j=1uij



n:ð22Þ

uij =1

k=1

dij

dkj

2

n-1

:ð23Þ

The basic steps of the FCM cluster algorithm are as follows under these two

conditions:

1. The number of clusters cis given, 1 <c≤m, and mis the number of samples.

Given that Tis the maximum number of iterations, εis the threshold, and ωis the

fuzzy number; the indicator setting iterative counter t=0.

2. Rectify partition matrix U

(t)

using Eq. (21).

3. Obtain the new cluster center c

(t) using Eq. (20).

4. t←t+ 1; repeat steps 2 and 3 until t≥Tor|U

(t)

-U

(t-1)

|≤ε.

3 Empirical Analysis

3.1 Sample Selection and Data Sources

This paper uses credit statistics on 687 small retail and wholesale enterprises,

representing customers of a Chinese commercial bank, to validate the model devel-

oped in Sect. 2. Further details about the credit rating indicators and default status of

these 687 small wholesale and retail enterprises are as follows. We select credit

rating indicators ﬁrst using the standard variables of rating agencies such as Standard

& Poor, Moody, and Fitch (Standard and Pool’s Services, 2011; Fitch Ratings, 2013;

Dagong, 2010), and second from papers on credit rating (Mijid & Bernasek, 2013;

Hai et al., 2013; Shi & Chi, 2014; Shi et al., 2016; Abedin et al., 2018,2019a,b; Sun

et al., 2022). That is, a total of 107 indicators are selected on repayment ability and

willingness to repay, and so forth. These indicators cover seven secondary standard

layers such as ﬁnancial factors, non-ﬁnancial factors, and the personal situation of

the legal representative of small wholesale and retail enterprises. Furthermore, we

eliminated 26 indicators for which statistics are unavailable, leaving 81 indicators, as

shown in Table 1.

3.2 Credit Rating of Small Wholesale and Retail Enterprises

1. Establishment of a Credit Risk Evaluation Indicator System

The original and standardized data on 687 small enterprises are shown in Tables 2

and 3, respectively.

(1) No. (5) Indicators (6) Type result

Assessing and Predicting Small Enterprises’Credit Ratings:... 137

Table 1 Screening criteria for indicators of small enterprise credit rating

(2) First

criterion

level

(3) Second

criterion

level

(4) Third

criterion

level

(7) Screening

1 Repayment

ability

Financial

factors

Solvency Debt asset

ratio

Negative Probit delete

... ... ... ...

28 Source of

repayment

Qualitative Unobservable

... ... ... ... ...

55 Growth

capacity

Revenue

growth

Positive Pass

... ... ... ...

63 Wages, wel-

fare growth

rate

Positive Unobservable

64 External macroeco-

nomic conditions

Industry sen-

timent index

Positive Pass

... ... ... ...

72 Economic

environment

Qualitative Unobservable

73 Internal non-ﬁnancial

factors

Years of rel-

evant

industry

Qualitative Probit delete

... ... ... ...

86 Willingness

to repay

Legal person situation Education

background

Qualitative Pass

... ... ... ...

98 Owner

qualities

Qualitative Unobservable

99 Enterprise credit

situation

Registered

capital

classiﬁcation

Qualitative Partial corre-

lation analysis

delete

... ... ... ...

103 Commercial reputation Tax records Qualitative Partial corre-

lation analysis

delete

104 Legal

disputes

Qualitative Probit delete

... ... ... ...

106 No. of

breaches of

contract

Qualitative Probit delete

107 Pledge guarantee factor Mortgage/

pledge/

guarantee

Qualitative Probit delete

138 B. Shi

Table 2 Original data for a sample of small retail and wholesale enterprises

(a) No. (b) Criterion level (c) Indicators

Original data

681 non-defaulting enterprises 6 defaulting enterprises

(1) C001 ... C681 (682) C682 ... (687) C687

internal

non-ﬁnancial

factors

years of relevant industry 8 ... 10 8 ... 10

... ... ... ... ... ... ... ...

10 C

legal person

situation

education background Junior diploma ... Bachelor’s

degree

N/A ... Bachelor’s degree

... ... ... ... ... ... ...

20 X

the value of car and real

estate of legal representatives

1000 ... 1000 N/A ... 100

21 C

Enterprise

credit situation

registered capital

classiﬁcation

Found ... Found 0.917 ... 0.917

... ... ... ... ... ... ... ... ...

27 C

operating

capacity

accounts receivable

turnover rate

5.00 ... 13.19 0 ... 9.17

... ... ... ... ... ... ... ...

36 X

cash conversion cycle -3973.69 ... 7.50 N/A ... 2.72

37 C

proﬁtability X

rate of return on common

stockholders’equity

0.078 ... 0.003 0.000 ... 0.280

... ... ... ... ... ... ... ...

49 X

operating activities gen-

erate cash inﬂows

112,458,001 ... 625,800,630 0.000 ... 26,139,847.75

50 C

growth capacity X

operating income growth

rate

0.000 ... 0.023 0.00 ... 1.36

... ... ... ... ... ... ... ...

54 X

retained revenue growth

rate

0.076 ... 1.251 0.510 ... 0.507

55 C

solvency X

debt asset ratio 6.84 ... 0.56 0 ... 0.604

... ... ... ... ... ... ... ...

74 X

EBITDA/total debt ratio 0.043 ... 0.003 -0.04 ... 0.49

75 C

external macro-

economic

conditions

industry sentiment index 137.45 ... 139.50 137.45 ... 127.20

... ... ... ... ... ... ... ...

80 X

Engel coefﬁcient 39.4 ... 37.0 39.40 ... 37.90

81 C

pledge guaran-

tee factor

mortgage/pledge/

guarantee

The guarantee

amount is 5 million

yuan

... No

guarantee

The guarantee

amount is 18.9 mil-

lion yuan

... The guarantee

amount is 3 million

yuan

82 —— Default 0 ... 1... 1

Assessing and Predicting Small Enterprises’Credit Ratings:... 139

(a)

No.

(b) Criterion

level (c) Indicator

140 B. Shi

Table 3 Standardized data

Standardized Data

681 non-default

enterprises

6 default

enterprises

C001 ... C681 C682 ... C687

internal

non-ﬁnancial

factors

Years of relevant industry 0.917 ... 0.917 0.917 ... 0.083

... ... ... ... ... ... ... ...

the proportion of total

amount of money returned

by enterprises through the

bank

0.667 ... 1.000 0.000 ... 0.000

... ... ... .. . ... ... ... . .. ...

81 C

pledge

guarantee

factor

mortgage/pledge/

guarantee

0.650 ... 0.000 0.000 ... 0.700

82 —— Default 0 ... ... 1

Taking C1 enterprise’s internal non-ﬁnancial factors as an example, the process

of partial deleting correlation indicator is illustrated (see Table 3). We put data on

nine indicators related to “internal non-ﬁnancial factors at enterprise C1”in Table 3

into Eqs. (3)–(6), so as to calculate r

, the partial correlation coefﬁcient of the

indicators. We respectively calculate the F-statistic of the indicator pairs whose

partial correlation coefﬁcients are over 0.7. Then we delete an indicator with a

smaller F-statistic and retain the other one. The result is shown in Table 4. The

rest can be done in the same manner. Using PCA, this paper removes 14 indicators

with redundant information.

After deleting some indicators with PCA, we screen the remaining indicators in

all standard layers through probit regression, and select the indicators with remark-

able discriminatory power on defaulting status. Then we put the remaining 67 indi-

cator data screened by partial correlation in Table 3into Eqs. (7)–(9) and screen them

using Stata. The 17 remaining screened indicators are in Table 5.

2. Solution to Credit Scoring of Small Wholesale and Retail Enterprises

The weight of 17 variables is calculated by the entropy weight in Table 5. With

Eqs. (10)–(13), it is easy to calculate the credit scores of the enterprises. The result is

presented in Table 6.

Then, we put the credit scores of these enterprises in Eqs. (14)–(16) and subse-

quently obtain the prediction accuracy of 85.40%. The result of the model classiﬁ-

cation is presented in Table 7, and the corresponding ROC curve is presented in

Fig. 6, where the area under ROC curve (AUC) is 0.909, suggesting the strong

predictive accuracy of the defaulting status of small enterprises obtained using the

screened 17 indicators.

Assessing and Predicting Small Enterprises’Credit Ratings:... 141

Table 4 Partial correlation deletion indicator related to “Internal non-ﬁnancial factors”

(1) No.

Indicators with a partial correlation coefﬁcient greater than 0.7 (6) Partial

correlation

coefﬁcient (7) Deleted indicator

(2) Indicator 1

(3) F-statistic of

indicator 1 (4) Indicator 2

(5) F-statistic

of indicator 2

debt asset ratio 2.370 X

shareholder equity ratio 2.392 0.993 X

debt asset ratio

current liabilities

operating ratio

1.284 X

Total debt operating

activity net cash ﬂow ratio

0.907 0.967 X

Total debt operating

activity net cash ﬂow ratio

quick ratio 0.079 X

cash ratio 0.753 0.809 X

cash ratio

(a) No. (b) Indicators Weight ...

Actual default status

142 B. Shi

Table 5 Credit indicators weights for small wholesale and retail enterprises

(c)

Standardized data

(1)

C001

(687)

C687

education background 0.025 0.500 ... 0.700

gender 0.003 1.000 ... 1.000

age 0.006 0.970 ... 0.848

family monthly income 0.172 0.071 ... 0.071

time in current position 0.047 0.250 ... 0.250

the value of car and real estate of legal

representatives

0.095 0.917 ... 0.917

ﬁx capital ratio 0.197 0.003 ... 0.029

operating income growth rate 0.033 0.197 ... 0.201

proﬁt growth rate 0.001 0.494 ... 0.530

10 X

Total asset growth rate 0.027 0.271 ... 0.298

11 X

capital accumulation rate 0.001 0.496 ... 0.496

12 X

retained revenue growth rate 0.017 0.510 ... 0.518

13 X

Industry sentiment index 0.001 0.633 ... 0.833

14 X

per capita disposable income of urban and rural

residents at the end of the year

0.001 0.300 ... 0.002

15 X

residential price index 0.000 0.817 ... 0.988

16 X

per capita disposable income of urban residents 0.007 0.155 ... 1.000

17 X

Engel coefﬁcient 0.001 0.576 ... 0.821

Table 6 Credit scoring of small enterprises

(1) No. (2) Loan No. (3) Original credit score P

(4) Standardized credit score S

1 200410270004 0.391 48.846

2 200412150123 0.243 0.759

... ... ... ...

687 X2012060800099 0.453 89.149

Table 7 Classiﬁcation of

credit rating system Model prediction result

1 (Default) 0 (Non-default) Sum

1 (default) 4 2 6

0 (non-default) 96 585 681

Sum 100 587 687

3. Credit Rating of Small Wholesale and Retail Enterprises

According to credit rating procedures, ﬁrst we set the number of credit rating

clusters to 9; the maximum number of iterations T=1000; the threshold

ε=1E-5; and the fuzzy number ω=2 (Zhong et al., 2014; Robillard et al.,

2014). Then, we use the vector S

of credit scores in MATLAB to get the

corresponding data distribution and classiﬁcation into clusters, as shown in

Figs. 7and 8; the changing trends in the objective functions are shown in

Fig. 9. Finally, the credit scores of cluster centers are presented in Table 8to

obtain nine corresponding ratings (AAA, AA, ..., C). Using the upper and

lower limits of credit scores, the credit score intervals can be obtained for

customers in different clusters (Table 8).

Assessing and Predicting Small Enterprises’Credit Ratings:... 143

Fig. 6 ROC curve

(AUC =0.909)

0100 200 300 400 500 600 700

100

Namber of cases

erocstiderC

Fig. 7 Distribution of credit score data for 687 small wholesale and retail enterprises

144 B. Shi

020 40 60 80 100 120

100

Namber of cases

erocstiderC

Fig. 8 The classiﬁcation of nine cluster centers

Fig.9 The changing trend

of credit rating division

objective function

010 20 30 40 50 60 70

0.5

1.5

2.5 x 10

Iteration

eulavdlohserhT

4 Conclusion

Small and medium-sized enterprises are important for the economic development of

China. However, because of imperfect ﬁnancial information, urgent demand for

loans but small amount of loan business, dispersed risks, and the absence of

necessary guarantees, small enterprises have made it difﬁcult for ﬁnancial institu-

tions such as commercial banks to depict their credit risks precisely, thus bringing

about salient loan difﬁculties in terms of ﬁnancing and high loan prices. This paper

(1) No.

uses a sample of 687 small enterprises to develop a credit rating system for these

enterprises using a combination of metrological statistics and fuzzy decision. To

begin with, we use partial correlation analysis to eliminate indicators with repeated

information and Probit regression to screen indicators that markedly inﬂuence the

defaulting status of small enterprises, establishing a credit risk evaluation indicator

system composed of 17 indicators such as “X18 family monthly income”and “X20

the value of car and real estate of legal representatives”for these enterprises. Second,

the credit scores of loan enterprises are calculated using the entropy-weighting

TOPSIS method. Finally, a fuzzy C-means (FCM) algorithm is used to evaluate

the credit ratings of small enterprises. The proposed system, through defaulting state

testing, shows the predictive accuracy of 85.40% and 90.09%, respectively,

conﬁrming a high default predictive capacity, which can be useful for commercial

banks.

Assessing and Predicting Small Enterprises’Credit Ratings:... 145

Table 8 The credit rating for small enterprises

(2) Cluster center of credit

score

(3) Credit

rating

(4) Credit score

interval

(5) Number of

cases

1 85.497 AAA [80.447, 100] 32

2 74.423 AA [71.347, 80.447) 60

3 68.251 A [65.264, 71.347) 54

4 62.147 BBB [59.232, 65.264) 68

5 56.153 BB [53.468, 59.232) 120

6 50.746 B [47.179, 53.468) 73

7 43.464 CCC [39.083, 47.179) 79

8 34.279 CC [27.826, 39.083) 68

9 19.883 C [0, 27.826) 124

This study is innovative in the following three respects. Firstly, the study pro-

poses a credit rating system consistent with the credit characteristics of small retail

and wholesale enterprises. It is an effective complement to existing credit rating

literature and can act as a decision-making reference for commercial banks and small

wholesale and retail enterprises in their credit rating. Second, triangular fuzzy

numbers are introduced into the scoring process, leading to the objective arbitrari-

ness. Third, the empirical research in this study shows that, for small retail and

wholesale enterprises, non-ﬁnancial indicators are more important for the prediction

of default risks than ﬁnancial factors. According to Fig. 5, among the 17 inﬂuential

rating indicators, the sum of the weights of non-ﬁnancial factors and external micro

indicators is 0.752, which is much higher than 0.248, the weight of internal ﬁnancial

indicators. Thus, non-ﬁnancial factors and external microeconomic conditions are

more important factors in inﬂuencing small and medium-sized wholesale and retail

credit ratings; non-ﬁnancial factors should be investigated in terms of the prediction

of small enterprises’default.

The study progressed in the development of credit rating systems for small

wholesale and retail companies, but there were still some limitations. Due to the

difﬁculty of getting real default losses data from loan companies, this paper uses

default status y

only as a dependent variable. This rating method has difﬁculty in

explaining the objective reality that two different customers who default at the same

time cause different losses to the same bank. With the accumulation of default data

and the advance of data analysis technology, further breakthroughs and research on

these problems can be produced.

146 B. Shi

Acknowledgments The study was supported by the National Natural Science Foundation of

China (Nos: 71873103, 72173096, 71503199 and 71731003), the Social Science Foundation of

Shaanxi Province, China (No. 2018D51), the Tang Scholar Program of Northwest A&F University,

China (No. 2021-04).

References

Abedin, M. Z., Chi, G. T., Colombage, S., & Moula, F. E. (2018). Credit default prediction using a

support vector machine and a probabilistic neural network. Journal of Credit Risk, 14(2), 1–27.

Abedin, M. Z., Chi, G. T., Moula, F. E., Zhang, T., & Hassan, M. K. (2019a). An optimized support

vector machine intelligent technique using optimized feature selection methods: Evidence from

Chinese credit approval data. Journal of Risk Model Validation, 13(2), 1–46.

Abedin, M. Z., Guotai, C., Moula, F. E., Azad, A. S. M. S., & Khan, M. S. U. (2019b). Topological

applications of multilayer perceptrons and support vector machines in ﬁnancial decision support

systems. International Journal of Finance & Economics, 24, 474–507.

Abedin, M. Z., Guotai, C., Uddin, M. M., Shahriare, S., Imran, K., & Hajek, P. (2021). Tax default

prediction using feature transformation-based machine learning. IEEE Access, 9, 19864–19881.

Abedin, M. Z., Chi, G. T., Hajek, P., & Tong, Z. (2022). Combining weighted SMOTE with

ensemble learning for the class-imbalanced prediction of small business credit risk. Complex &

Intelligent Systems.https://doi.org/10.1007/s40747-021-00614-4

Akkoç, S. (2012). An empirical comparison of conventional techniques, neural networks and the

three-stage hybrid adaptive neuro fuzzy inference system (ANFIS) model for credit scoring

analysis: The case of Turkish credit card data. European Journal of Operational Research,

222(1), 168–178.

Altman, E. I. (1968). Financial ratios, discriminant analysis and the prediction of corporate

bankruptcy. Journal of Finance, 23(4), 589–609.

Altman, E. I., Haldeman, R., & Narayanan, P. (1977). ZETA analysis: A new model to identify

bankruptcy risk of corporations. Journal of Banking and Finance, 1,29–54.

Bahnsen, A. C., Aouada, D., & Ottersten, B. (2015). Example-dependent cost-sensitive decision

trees. Expert Systems with Applications, 42(19), 6609–6619.

Bai, X. P., & Zhao, Z. C. (2022). An optimal credit scoring model based on the maximum default

identiﬁcation ability for Chinese small business. Discrete Dynamics in Nature and Society,

2022, 1551937.

Bai, C. G., Shi, B. F., Liu, F., & Joseph, S. (2019). Banking credit worthiness: Evaluating the

complex relationships. Omega, 83,26–38.

Bank of Dalian. (2014). Dalian Bank small business credit system. Dalian Bank.

Chai, N. N., Wu, B., Yang, W. W., & Shi, B. F. (2019). A multicriteria approach for modeling small

enterprise credit rating: Evidence from China. Emerging Markets Finance and Trade, 55(11),

2523–2543.

Cheng, J., Feng, Y., Tan, J., & Wei, W. (2008). Optimization of injection mold based on fuzzy

moldability evaluation. Journal of Materials Processing Technology, 208(1), 222–228.

Assessing and Predicting Small Enterprises’Credit Ratings:... 147

Chern, C. C., Lei, W. U., Huang, K. L., et al. (2021). A decision tree classiﬁer for credit assessment

problems in big data environments. Information Systems and e-Business Management, 19,

363–386.

Chi, G. T., & Zhang, Z. (2017). Multi criteria credit rating model for small Enterprise using a

nonparametric method. Sustainability, 9(10), 1–23.

Chi, G. T., Abedin, M. Z., & Moula, F. E. (2017). Modeling credit approval data with neural

networks: An experimental investigation and optimization. Journal of Business Economics and

Management, 18(2), 224–240.

China Banking Regulatory Commission (CBRC). (2015). Supervision by law, supervision of the

people, risk supervision: The China Banking Regulatory Commission implements the reform of

the regulatory framework. Accessed from http://www.cbrc.gov.cn/chinese/home/docView/671

63D0D8293499BA499D2A9705C61CD.html

Ciampi, F., & Gordini, N. (2013). Small Enterprise default prediction modeling through artiﬁcial

neural networks: An empirical analysis of Italian small enterprises. Journal of Small Business

Management, 51(1), 23–45.

Dagong Global Credit Rating Co. (2010). Credit rating methodology framework of Dagong Global

Credit Rating Co., Ltd.

Demircan, S., & Kahramanli, H. (2016). Application of fuzzy C-means clustering algorithm to

spectral features for emotion classiﬁcation from speech. Neural Computing & Applications,

29(8), 1–8.

Fitch Ratings. (2013). Fitch Ratings global corporate ﬁnance 2012 transition and default study.

Credit Market Research, Fitch Ratings.

Florez-Lopez, R. (2007). Modelling of insurers’rating determinants: An application of machine

learning techniques and statistical models. European Journal of Operational Research, 183(3),

1488–1512.

Florez-Lopez, R., & Ramon-Jeronimo, J. M. (2015). Enhancing accuracy and interpretability of

ensemble strategies in credit risk assessment: A correlated-adjusted decision forest proposal.

Expert Systems with Applications, 42(13), 5737–5753.

Gu, W., Meheli, B., Zhang, C., & Li, R. W. (2017). A uniﬁed framework for credit evaluation for

internet ﬁnance companies: Multi-criteria analysis through AHP and DEA. International Jour-

nal of Information Technology & Decision Making, 3(3), 597–624.

Hai, L., Shi, B. F., & Peng, G. (2013). A credit risk evaluation index system establishment of petty

loans for farmers based on correlation analysis and signiﬁcant discriminant. Journal of Soft-

ware, 8(9), 2344–2351.

Harris, T. (2015). Credit scoring using the clustered support vector machine. Expert Systems with

Applications, 42(2), 741–750.

Hens, A. B., & Tiwari, M. K. (2012). Computational time reduction for credit scoring: An

integrated approach based on support vector machine and stratiﬁed sampling method. Expert

Systems with Applications, 39(8), 6774–6781.

İç, Y. T., & Yurdakul, M. (2010). Development of a quick credibility scoring decision support

system using fuzzy TOPSIS. Expert Systems with Applications, 37(1), 567–574.

Industrial and Commercial Bank of China. (2005). Notice on printing and distributing the measures

for the evaluation of credit Ratings of small business Enterprises of Small Industrial Enterprises

of Industrial and Commercial Bank of China. Industrial and Commercial Bank of China, no. 78.

iResearch. (2021). Report on Chinese Financing Development of Micro, Medium and Small

Enterprises in 2021. iResearch Inc.

Khalili-Damghani, K., Sadi-Nezhad, S., Lotﬁ, F. H., & Tavana, M. (2013). A hybrid fuzzy rule-

based multi-criteria framework for sustainable project portfolio selection. Information Sciences,

220(1), 442–462.

Liang, X. C., Chen, S. F., & Liu-Yan. (2007). The study of small enterprises credit evaluation based

on incremental AntClust. In IEEE International Conference on Grey Systems and Intelligent

Services (pp. 294–298). IEEE Xplore.

148 B. Shi

Lu, Y., Yang, L., Shi, B., Li, J., & Abedin, M. Z. (2022). A novel framework of credit risk feature

selection for SMEs during industry 4.0. Annals of Operations Research, 2022,1–28.

Marcano-Cedeño, A., Marin-De-La-Barcena, A., Jimenez-Trillo, J., Piñuela, J. A., & Andina,

D. (2011). Artiﬁcial Metaplasticity neural network applied to credit scoring. International

Journal of Neural Systems, 21(4), 311–317.

Mijid, N., & Bernasek, A. (2013). Gender and the credit rationing of small businesses. Social

Science Journal, 50(1), 55–65.

Nami, S., & Shajari, M. (2018). Cost-sensitive payment card fraud detection based on dynamic

random forest and K-nearest neighbors. Expert Systems with Applications, 110, 381–392.

Promentilla, M. A., Furuichi, T., Ishii, K., & Tanikawa, N. (2008). A fuzzy analytic network

process for multi-criteria evaluation of contaminated site remedial countermeasures. Journal of

Environmental Management, 88(3), 479–495.

Reichert, A., Cho, C. C., & Wagner, G. (1983). An examination of the conceptual issues involved in

developing credit-scoring models. Journal of Business & Economic Statistics, 1(2), 101–114.

Robillard, M. P., Maalej, W., Walker, R. J., & Zimmermann, T. (2014). An approach on fault

detection in diesel engine by using symmetrical polar coordinates and image recognition.

Advances in Mechanical Engineering, 2014(4), 1–9.

Ruan, J. H., Wang, Y., Chan, F. T. S., Hu, X., Zhao, M., Zhu, F., Shi, B. F., Shi, Y., & Lin,

F. (2018). A life-cycle framework of green IoT based agriculture and its ﬁnance, operation and

management issues. IEEE Communications Magazine, 2018.https://doi.org/10.1109/MCOM.

2018.1800332

Rui, L., & Mendes, R. V. (2017). Detecting and quantifying ambiguity: A neural network approach.

Soft Computing, 22(8), 1–9.

Shi, B. F., & Chi, G. T. (2014). A model for recognizing key factors and applications thereof to

engineering. Mathematical Problems in Engineering, 2014(1), 368–381.

Shi, B. F., Wang, J., Qi, J., & Cheng, Y. (2015). A novel imbalanced data classiﬁcation approach

based on logistic regression and ﬁsher discriminant. Mathematical Problems in Engineering,

2015(6), 1–12.

Shi, B. F., Chen, N., & Wang, J. (2016). A credit rating model of microﬁnance based on fuzzy

cluster analysis and fuzzy pattern recognition: Empirical evidence from Chinese 2,157 small

private businesses. Journal of Intelligent & Fuzzy Systems, 31(6), 3095–3102.

Shi, B. F., Meng, B., Yang, H. F., Wang, J., & Shi, W. L. (2018). A novel approach for reducing

attributes and its application to small enterprise ﬁnancing ability evaluation. Complexity, 2018,

1–17. https://doi.org/10.1155/2018/1032643

Sobehart, J., & Keenan, S. (2001). Measuring default accurately. Risk, 14(3), 31–33.

Standard & Poor’sRatings Services. (2011). S&P’s study of China’s top corporates highlights their

signiﬁcant ﬁnancial risks. Standard & Poor’s.

State Council of the People’s Republic of China (SCPRC). (2016). Notice of the State Council on

Printing and Promoting Inclusive Financial Development Plan (2016–2020). Accessed from

http://www.gov.cn/zhengce/Content/2016–01/15/content_10602.htm

State Council of the People’s Republic of China (SCPRC). (2017). Notice of the General Ofﬁce of

the State Council on the Establishment of the “Made in China 2025”National Demonstration

Zone. Accessed from http://www.gov.cn/zhengce/content/201711/23/content_5241727.htm

Sun, Y., Chai, N. N., Dong, Y. Z., & Shi, B. F. (2022). Assessing and predicting small industrial

enterprises’credit ratings: A fuzzy decision making approach. International Journal of Fore-

casting, Forthcoming., 38, 1158.

Tomczak, J. M., & Zięba, M. (2015). Classiﬁcation restricted Boltzmann machine for comprehen-

sible credit scoring model. Expert Systems with Applications, 42(4), 1789–1796.

Wang, Y.T. and Leng, H. Y, 2021. Credit decision of SMEs based on improved TOPSIS and

decision tree. In 2021 4th International Conference on Advanced Electronic Materials, Com-

puters and Software Engineering (AEMCSE) (pp. 489–492). https://doi.org/10.1109/

AEMCSE51986.2021.00106

Assessing and Predicting Small Enterprises’Credit Ratings:... 149

Wang, J., Ding, D., Liu, O., & Li, M. (2016). A synthetic method for knowledge management

performance evaluation based on triangular fuzzy number and group support systems. Applied

Soft Computing, 39,11–20.

Wu, Z., Ahmad, J., & Xu, J. (2016). A group decision making framework based on fuzzy VIKOR

approach for machine tool selection with linguistic information. Applied Soft Computing, 42,

314–324.

Yu, F., Xu, H., Wang, L., & Zhou, X. (2010). An improved automatic FCM clustering algorithm. In

International Workshop on Database Technology and Applications (pp. 1–4). IEEE. https://doi.

org/10.1109/DBTA.2010.5659043

Yurdakul, M., & Iç, Y. T. (2015). Development of a performance measurement model for

manufacturing companies using the AHP and TOPSIS approaches. International Journal of

Production Research, 43(21), 4609–4641.

Zhang, J. H., Liu, H. Y., Zhu, R., & Liu, Y. (2017). Emergency evacuation of hazardous chemical

accidents based on diffusion simulation. Complexity, 2017,1–16.

Zhong, Y., Zhang, L., Xing, S., Li, F., & Wan, B. (2014). The big data processing algorithm for

water environment monitoring of the three gorges reservoir area. Abstract and Applied Analysis,

2014(5), 1–7.

Zhu, P., & Hu, Q. (2013). Rule extraction from support vector machines based on consistent region

covering reduction. Knowledge-Based Systems, 42(2), 1–8.

Part III

Financial Time-Series Forecasting

153

An Ensemble LGBM (Light Gradient

Boosting Machine) Approach for Crude Oil

Price Prediction

Sad Wadi Sajid, Mahmudul Hasan, Md. Fazle Rabbi,

and Mohammad Zoynul Abedin

Abstract Crude oil is considered one of the most important resources in the world

today. Most of the fuel used today is reﬁned from crude oil. Fuel also has a great

impact on the global economy. The crude oil market is liquid and uncertain. The

prediction of the crude oil market price has become a necessity of every second for

governments, industries, and individuals. Predicting the price of crude oil can help to

achieve a sustainable economy. The goal of this study is to forecast crude market

prices as accurately as possible using machine learning and ensemble learning

methodology. In this study, we propose the prediction of crude oil using Light

Gradient Boosting (LGBM), Random Forest ensemble machine learning algorithm,

Lasso Regression, and Decision Tree machine learning algorithm. The BRENT time

series crude oil data are used for analysis and form a prediction model that gives less

error and more accuracy. We have compared the prediction result of LBGM with

Lasso Regression, Random Forest Regression, and Decision Tree regression analy-

sis. A comparison curve is used for introducing the result, turns out LBGM gives the

most accurate and efﬁcient prediction result. We have validated our result by

evaluating the root mean square error (RMSE), mean absolute percentage error

(MAPE), mean squared error (MSE), mean absolute error (MAE), and the results

obtained by the proposed model are signiﬁcantly close and superior.

S. W. Sajid

Department of Electronics and Communication Engineering, Hajee Mohammad Danesh

Science and Technology University, Dinajpur, Bangladesh

M. Hasan · M. F. Rabbi

Department of Computer Science and Engineering, Hajee Mohammad Danesh Science and

Technology University, Dinajpur, Bangladesh

e-mail: rabbi@hstu.ac.bd

M. Z. Abedin (✉)

Department of Finance, Performance and Marketing, Teesside University International Business

School, Teesside University, Middlesbrough, Tees Valley, UK

e-mail: m.abedin@tees.ac.uk

©The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

M. Z. Abedin, P. Hajek (eds.), Novel Financial Applications of Machine Learning

and Deep Learning, International Series in Operations Research & Management

Science 336, https://doi.org/10.1007/978-3-031-18552-6_9

Keywords Crude oil price prediction · Ensemble learning · Machine learning ·

Time series analysis

154 S. W. Sajid et al.

1 Introduction

Crude oil is essentially the primary resource of major oils and fuels available today.

Crude oil is a type of petroleum. It is composed of natural hydrocarbon deposits and

other organic materials. Crude oil is found as a liquid substance in underground

reservoirs (Ashour et al., 2011). If the crude oil price increases generally the

expected rate of economic growth decreases. This essentially lowers the economic

growth prospects, in turn, decreases the expected earnings of companies, resulting in

a dampening effect on stock prices. Rather than that, volatilities in the price of crude

oil have a huge impact on other economic activities too, as crude oil is the largest

source in the energy market (Zhao et al., 2017). Oil price prediction is very useful

and important for companies, industries, researchers, governments, and individuals.

Because crude oil has a great impact on the world economy and stability (Chen &

Huang, 2021; Abedin et al., 2021a). Like the FOREX market, the crude oil market is

very volatile, so it has been an interesting ﬁeld for researchers. There are already

many methods that have been developed to predict crude oil prices. Many of them

use convolutional neural networks (CNN), artiﬁcial neural networks (ANN), deep

neural networks (DNN) (Abedin et al., 2021a,b; Rahman et al., 2021; Kaur et al.,

2013). From EIA we get to know that US oil production has increased from 4.96

million barrels per day to 5.59 million barrels per day in just the last ﬁve years.

OPEC’s recent agreement is causing volatility in the oil price. For this reason, the

environment of the oil market is changing and inﬂuencing factors are becoming

more and more complex and diverse (Lu et al., 2021). Therefore, forecasting the

price of crude oil has become more difﬁcult for researchers; they are applying new

and more efﬁcient approaches such as stream learning, CNN model, ANN, vector

autoregressive model, etc. (Chen & Huang, 2021; Abedin et al., 2021a; Rahman

et al., 2021; Kaur et al., 2013). Authors of those study included many factors,

different approaches. Among all of them, “Ensemble Machine Learning”has been

shown to give the most desired result. Authors of this study intend to get the best

possible forecasting result; authors started with machine learning approaches Lasso

Regression, Decision tree regression and Bootstrap Aggregation (Bagging) ensem-

ble Random Forest Analysis. Both of them gave a good result, but why not analyze it

using a better and more efﬁcient forecasting system for the crude oil market. So, the

authors use the stochastic boosting ensemble model named “Light Gradient

Boosting Machine (LBGM),”which gives the best possible forecasting result.

Although authors have found that “Random Forest Analysis”provides better results

than “Lasso Regression”as the crude oil price is a nonlinear time series data. The

prediction model the authors have built is promising and it will provide an upcoming

ﬂuctuation in the price of crude oil. Different types of error measurement techniques

are used to measure the performance of the algorithms are shown in tabular format.

Also, the error is represented by a line chart that clearly indicates that the

performance of Light GBM is better than others. Later parts of this study have

reviewed on related work, methodology, performance measurement, result and

discussion, conclusion, and future work that the authors intend to do.

An Ensemble LGBM (Light Gradient Boosting Machine) Approach for Crude... 155

2 Literature Review

As already mentioned in recent years, many remarkable works have been done on

economic predictions. A study proposed a model based on Bidirectional Long Short-

Term Memory (Bi-LSTM) for oil price forecast. This proposed framework has two

modules (Vo et al., 2020). Zhongpei Chen approached the crude oil price prediction

method with Long Short-Term Memory (LSTM) deep learning. They proposed a

creative algorithm named data transfer with prior knowledge. The study has also

compared the price forecast performance with three other training models, but

LSTM gave the most desired result (Cen & Wang, 2019). A novel algorithm was

introduced by the authors of the study to predict the variation in the price of crude oil

of the West Texas Intermediate (WTI) which is based on soft computing. This study

implemented a simple but effective way to predict the price using a data ﬁltering

algorithm (Ghaffari & Zare, 2009). A novel network “Random Wavelet Neural

Network”combined with effective random time function is developed by the authors

to improve the prediction accuracy of ﬂuctuations in crude oil price. This study

predicted both WTI and BRE crude oil prices using a custom-developed model

(Huang & Wang, 2018). The prediction selection method was introduced, rather than

widely used regressors, resulting in improvements in prediction accuracy close to

10% relative to the benchmark. The authors pointed out that the well-known Welch

and Goyal’s dataset leads to more consistent and remarkable accuracy gains relative

to other alternative approaches (Nonejad, 2021; Welch & Goyal, 2008). Various

types of deep learning approaches have been applied to predict the exchange rate

during the COVID-19 pandemic, and the authors here worked with a few interesting

parameters to prioritize the effect of the pandemic on the economy (Shajalal et al.,

2021). LSTM and GRU are widely used recurrent neural networks that are used to

predict various phenomena. GA Busari has shown the comparison between

Adaboost-LSTM and Adaboost-GRU, and the empirical result of that study shows

that Adaboost-GRU performs better than Adaboost-LSTM in predicting the price of

crude oil (Busari & Lim, 2021). Predicting a phenomenon has always been a favorite

for researchers. There are many approaches to do so, but a combination of traditional

and modern artiﬁcial intelligence has been shown to provide more accurate and

efﬁcient results. The authors of this study have proposed a “Hybrid Model”to

predict credit risk (Chi et al., 2019). Complex and volatile ﬁnancial markets are

well suited to gray analysis environments. So, the authors proposed a gray prediction

model that signiﬁcantly improved performance (Norouzi & Fani, 2020). Yanhui

Chan proposed a new deep learning-based hybrid crude oil price prediction model,

which improved the forecasting accuracy of previous works (Chen et al., 2017). The

more accurate oil price can be predicted, the more stable the market will

be. Real-time prediction is rare in the case of crude oil price forecasting; Yuan Zhao

proposed a new hybrid model that can provide online real-time price prediction

(Zhao et al., 2021). As the crude oil market is highly volatile, it is like an imbalance

of time series data. A novel ensemble approach was suggested by the authors to

predict an imbalance dataset (Abedin et al., 2019). Yifan Yang found that divide-

and-conquer strategy gains a better prediction performance. They have come up with

a hybrid approach based on K-means + KPCA + KELM based (Yang et al., 2021).

Many researchers have worked on predicting the price of crude oil. Autoregressive

moving average (ARMA) models and vector autoregressive (VAR) models with

diverse data input each time (Kulkarni & Haidar, 2009). If the crude oil price data are

strongly nonlinear, then these nonlinear models can produce more accurate results

(Bashiri Behmiri & Pires Manso, 2013). On the crude oil market, the uncertainty of

the price is a factor, as the value depends on many parameters. The machine learning

method based on adaptive Cuckoo search algorithm (AGWOCS) is proposed to

predict the volatile market price of crude oil. The effectiveness of the proposed

system, daily and weekly Brent oil prices, are modeled as a case study (Wang et al.,

2020). Binrong Wu proposed a novel text-based and big-data-driven model, which

utilized a convolutional neural network (CNN) to automatically scrap crude oil news

updates. This case study collected 4837 and 3883 news headlines to develop a text-

based crude oil forecasting system (Wu et al., 2021). Based on this analysis, in this

study, we use one of the latest ensemble algorithms called light gradient boosting

machine (LGBM) to predict the price of crude oil.

156 S. W. Sajid et al.

3 Research Methodology

The traditional approach of machine learning analysis is used to predict the price of

crude oil. Data are preprocessed before being split into training and testing sets. We

randomly split the dataset into 80:20 ratio for training and testing data. The analysis

model is built by machine learning and ensemble algorithms trained by the training

data, and after training the predicted values come out using the testing values as

input. A block diagram of our proposed methodology is given in Fig. 1.

3.1 Dataset

The dataset that was used for the analysis is Crude Oil Prices: Brent –Europe data. It

is taken from the US Energy Information Administration. It releases as spot prices,

and units is Dollar per barrel. Data frequency is daily, but not seasonally adjusted. It

is a time series data from May 20, 1987 to September 10, 2021, and the total number

of observations is 8954. Figure 2represents the information about the dataset.

The price of crude oil was stable during the period 1987 to 2000. After this time,

the price increases by a rate. In 2008–2009 it was the maximum and then the price

goes down. At the time 2011 to 2015 the price was in a stable situation and after the

period it started falling. In 2020 the price of crude oil fell due to the Covid-19

pandemic. The situation is going to be good now and the price is also increasing. The

plot clearly indicates that there is a great impact of Covid-19 on the price of crude oil.

The above discussion indicates that market of crude oil is not fully stable. Many

variables are responsible for varying this price. The prediction of this market is really

hard and requires a special and deep analysis. The numerical description of the data

set is given in Table 1.

An Ensemble LGBM (Light Gradient Boosting Machine) Approach for Crude... 157

Fig. 1 Proposed methodology for predicting the price of crude oil

100

120

140

160

5/20/1987

5/20/1989

5/20/1991

5/20/1993

5/20/1995

5/20/1997

5/20/1999

5/20/2001

5/20/2003

5/20/2005

5/20/2007

5/20/2009

5/20/2011

5/20/2013

5/20/2015

5/20/2017

5/20/2019

5/20/2021

Oil Price

Date

Crude Oil Price Data from 1987 to 2021

Fig. 2 Representation of crude oil price data from 1987 to 2021

The standard deviation of the crude oil price is 32.01776, and it is not too many

scatters. The price of crude oil is increasing day by day and is maintaining a rate. But

in the last three months of 2008 the price of crude oil was the highest, because the

stock was primarily caused by physical disruptions of supply and the strong demand

facing stagnating world production (Ratti & Vespignani, 2013).

158 S. W. Sajid et al.

Table 1 Descriptive statistics

of Brent Crude Oil data Mean Standard Deviation Min Max

46.75337 32.01776 9.10000 143.95000

3.2 Description of the Algorithms Used in Analysis

Two ensemble machine learning algorithms named Light Gradient Boosting and

Random Forest Regression as well as Lasso and Decision Tree machine learning

algorithm, are used for this analysis. The short description of the algorithms is given

below.

Lasso Regression The lasso is a type of linear regression and it is a shrinkage

method like a ridge. There is a little difference between them. LASSO stands for

Least Absolute Shrinkage and Selection Operator. The cost function for the lasso

regression can be deﬁned as follows:

i=1

yi-b

ðÞ

2=X

i=1

yi-X

j=0

wj×xij

þλX

j=0

for some t>0, X

j=0

wj

<t,ð1Þ

The main difference between the ridge and the Lasso regression cost function

equation is that magnitudes are considered in the Lasso regression instead of the

square coefﬁcient. This normalization (L1) can result in zero coefﬁcients, i.e., some

properties are completely ignored for output evaluation. As a result, Lasso regression

not only reduces overﬁtting, but also helps select features that facilitate the interpre-

tation of models.

Random Forest Random Forest is an ensemble classiﬁer that creates a number of

separate and non-identical decision trees using randomization (Datta et al., 2021).

This algorithm, which is a mixture of tree predictors, is used for both classiﬁcation

and regression. Each decision tree includes a random vector as a parameter, deter-

mines the feature of the samples at random, and chooses the training data set at

random from either a subset of the data set or the entire data set (Bradter et al., 2013).

The error rates are comparable to Ad boost when a random selection of features is

employed to divide each node, but they are more resilient in terms of turbulence

(Shakoor et al., 2017). Random Forest is a very ﬂexible and simple machine learning

technique that, in most cases, gives excellent results even without hyper-parameter

adjustment. Based on our need, we employed Random Forest for the regression

portion of our technique in this study. Utilizing random forest regression, we were

able to get very high accuracy for our dataset. SK-learn offers a useful tool for this

that quantiﬁes the signiﬁcance of a feature by looking at how much error is reduced

on all trees in the forest by tree nodes using that feature (Grange & Hand, 1987).

Overﬁtting is a problem with deep decision trees; however, overﬁtting is rarely a

problem with Random Forest. It generates random subsets of the characteristics and

uses these selections to form smaller trees that it then merges.

An Ensemble LGBM (Light Gradient Boosting Machine) Approach for Crude... 159

Fig. 3 Leaf-wise tree growth of Light Gradient Boosting Machine

Decision Tree Regression For supervised learning, a decision tree is a common

practical technique. It allows both classiﬁcation and regression estimates to be made.

The root node, inner node, and leaf node are the three types of nodes in a decision

tree, which is a tree-structured classiﬁer. The root node is the ﬁrst node, which

represents the entire sample and can be divided into other nodes. The core nodes

reﬂect the characteristics of the dataset, whereas the branches represent decision

rules. Finally, the root nodes represent the result. A decision tree is executed for a

speciﬁc data point, True/False questions are answered until they reach the leaf node.

The average value of the dependent variable at that particular leaf node is used to

produce the ﬁnal prediction. Through several iterations, the tree is able to predict an

appropriate value for the data point. Decision trees are useful because they are simple

to grasp, need minimal data cleansing, do not suffer from non-linearity, and have a

small number of hyper-parameters to tune.

Light Gradient Boosting Machine Light GBM is a tree-based learning algorithm-

based gradient boosting framework (Rufo et al., 2021). It is intended to be dispersed

and efﬁcient and provides the following advantages: reduced memory utilization,

increased training efﬁciency and speed, and better accuracy. This algorithm uses two

novel techniques called Gradient-Based One-Side Sampling (GOSS) and Exclusive

Feature Bunding (EFB), which makes it faster. The Light GBM approach is built on

a histogram that organizes continuous feature values into discrete bins to accelerate

the training process. Lower memory utilization: Continuous values are replaced with

discrete bins, resulting in lower memory usage. It makes this algorithm faster than

the others. The tree-based structure of this algorithm is given in Fig. 3.

3.3 Performance Measures

Machine learning and predictive analytics are indeed prone to a variety of errors. We

use four mostly used error measurement techniques and compare them using both

tabular and graphical forms. Here is a short overview of the errors with the

parameters:

160 S. W. Sajid et al.

MAE =1

i=1

yi -y

,MSE =1

nXn

i=1

y-yiðÞ

2,ð2Þ

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

RMSE =1

i=1

y-yiðÞ

2,MAPE =1

nXn

i=1

y-yiðÞ

y,ð3Þ

where nis the number of samples, Σis the summation symbol (which means “add

them all together”), yis the actual value, and yi is the predicted value.

MAE means Mean absolute error. Absolute errors are deﬁned as absolute values

that differ from prediction to actual values. MAE indicates the average error

expected from forecasts.

MSE means Mean square error. The average square error of the regression line

shows the distance to the point set. This is done by dividing the distance between

points and regression lines (these distances are the “errors”). Squaring is needed to

eliminate any negative signs.

RMSE means Root Mean Square Error, which is the standard deviation of the

residuals (prediction errors). Residuals are used to measure the distance between

data points and the regression lines; RMSE measures the distribution of these

residuals. In other words, it reveals how strongly the data is aggregated around the

line of best ﬁt.

MAPE means Mean absolute percentage error. One of the most widely used KPIs

for evaluating predictive performance is MAPE. MAPE is calculated by dividing the

total absolute error by the desired quantity (each period is separately). This is

calculated on an average percentage error.

4 Results and Discussion

In this paper, we use four different models to predict the price of crude oil. After

developing the model, we test by the test value and generate the actual vs. predicted

curve. The curves of the three methods are given in Figs. 4and 5.

Figure 4indicates the actual vs predicted curve of crude oil price prediction. The

blue color indicates the actual values and the red color indicates the predicted values

of the Brent oil price data. The curve shows that the performance of Lasso is good

and after evaluation we get 0.01730 MAE, 0.00046 MSE, 0.02143 RMSE and

0.40613 MAPE error, which are tabulated in Table 2. This curve also indicates

that the price of crude oil does not maintain any speciﬁc rules. It can fall at any time

and increase at any time. The results of the remaining algorithms are shown and

discussed below one by one.

Figure 5indicates the actual vs. predicted curve of crude oil price prediction. The

blue color indicates the actual values and the red color indicates the predicted values

of the Brent oil price data. The curve shows that the performance of Random Forest

An Ensemble LGBM (Light Gradient Boosting Machine) Approach for Crude... 161

Crude Oil Price Prediction Using Lasso Regression

0 250 500 750 1000 1250 1500 1750

Data Sample According to Time

0.0

0.2

0.1

0.3

0.4

0.5

Actual

Predicted

Crude Oil Price

Fig. 4 Actual vs. predicted using Lasso

Crude Oil Price Prediction Using Random Forest Regression

12501000 1500 17507505002500

Data Sample According to Time

0.0

0.1

0.2

0.3

0.4

0.5

Actual

Predicted

Crude Oil Price

Fig. 5 Actual vs. predicted using Random Forest

Table 2 Performance measurement of different algorithms for the prediction of crude oil price

Method MAE MSE RMSE MAPE

Lasso regression 0.01730 0.00046 0.02143 0.40613

Random Forest regression 0.01076 0.00020 0.01416 0.26699

Decision tree 0.01065 0.00019 0.01393 0.27218

Light gradient boosting 0.00732 0.00009 0.00998 0.26201

Bold values: the most minimum error rate that signiﬁes the best model performance

is good and after evaluation we get 0.01076 MAE, 0.00020 MSE, 0.01416 RMSE

and 0.26699 MAPE error, which are presented in Table 2.

162 S. W. Sajid et al.

Crude Oil Price Prediction Using Decision Tree

0 250 500 750 1000 1250 1500 1750

Data Sample According to Time

0.0

0.1

0.2

0.3

0.4

0.5

0.6 Actual

Predicted

Crude Oil Price

Fig. 6 Actual vs. predicted using Decision Tree

Figure 6indicates the actual vs. predicted curve of crude oil price prediction. The

blue color indicates the actual values and the red color indicates the predicted values

of the Brent oil price data. The curve shows that the performance of Decision Tree is

good and after evaluation we get 0.01065 MAE, 0.00019 MSE, 0.01393 RMSE and

0.27218 MAPE errors, which are presented in Table 2.

Figure 7indicates the actual vs. predicted curve of crude oil price prediction. The

blue color indicates the actual values and the red color indicates the predicted values

of the Brent oil price data. The curve shows that the performance of LGBM is good

and after evaluation we get 0.00732 MAE, 0.00009 MSE, 0.00998 RMSE, and

0.26201 MAPE error, which are tabulated in Table 2.

Table 2represents the MAE, MSE, RMSE, and MAPE error values of Lasso

Regression, Random Forest Regression, and Light Gradient Boosting. It clearly

indicates all kinds of error in Light Gradient Boosting are less than others. It

means that the prediction of Light Gradient Boosting is better than the other two

algorithms. For clear understanding, we represent the errors in a line chart in Fig. 7.

Figure 8represents MAE, MSE, RMSE, and MAPE of three models. The yellow

color represents the errors of the Light Gradient Boosting algorithm, the gray color

represents the Decision Tree, the orange color represents the errors of the Random

Forest Regression, and the blue color represents the errors of Lasso Regression. The

numeric values 1, 2, 3, and 4 represent MAE, MSE, RMSE, and MAPE consecu-

tively. The ﬁgure clearly indicates that the error rate of Light Gradient Boosting is

less than others.

An Ensemble LGBM (Light Gradient Boosting Machine) Approach for Crude... 163

Crude Oil Price Prediction Using Light Gradient Boosting

0 250 500 750 1000 1250 1500 1750

Data Sample According to Time

Actual

Predicted

0.0

0.1

0.2

0.3

0.4

0.5

0.6

Crude Oil Price

Fig. 7 Actual vs. predicted using LGBM

Fig. 8 Comparison of different methods by errors

5 Conclusion and Future Work

This research focuses on the prediction of Brent crude oil price. We apply two

machine learning algorithms and two ensemble algorithms for analysis. Overall

performance of Light Gradient Boosting Machine algorithms is better than others.

All the measurements are shown in both tabular and graphical form. The perfor-

mance of the other algorithms is also satisfying and error is low. This analysis helps

all those related to this ﬁeld take the challenging decisions that are directly and

indirectly depend on the price of crude oil.

164 S. W. Sajid et al.

In the future, we want to build an API that shows the prediction of crude oil real-

time price. The authors want to add more parameters to the input, and to minimize

the complexity of the space and time of the model to ensure accurate prediction. The

authors also want to prepare an application software that anyone can use to obtain the

real-time predictions.

References

Abedin, M.Z., Guotai, C. and Moula, F.E., 2019, June. Weighted SMOTE-ensemble algorithms:

Evidence from Chinese imbalance credit approval instances. In 2019 2nd International Con-

ference on Data Intelligence and Security (ICDIS) (pp. 208–211). IEEE Xplore.

Abedin, M. Z., Hasan, M. M., Hassan, M. K., & Hajek, P. (2021a). Deep learning-based exchange

rate prediction during the COVID-19. Annals of Operations Research,1–52.

Abedin, M. Z., Hassan, M. K., Khan, I., & Julio, I. F. (2021b). Feature transformation for corporate

tax default prediction: Application of machine learning approaches. Asia-Paciﬁc Journal of

Operational Research, 39, 2140017.

Ashour, I., Al-Rawahi, N., Fatemi, A., & Vakili-Nezhaad, G. (2011). Applications of equations of

state in the oil and gas industry. Thermodynamics Kinetics of Dynamic Systems, 1, 165–178.

Bashiri Behmiri, N., & Pires Manso, J. R. (2013). Crude oil price forecasting techniques: A

comprehensive review of literature. Available at SSRN 2275428.

Bradter, U., Kunin, W. E., Altringham, J. D., et al. (2013). Identifying appropriate spatial scales of

predictors in species distribution models with the random forest algorithm. Methods in Ecology

and Evolution, 4(2), 167–174.

Busari, G. A., & Lim, D. H. (2021). Crude oil price prediction: A comparison between AdaBoost-

LSTM and AdaBoost-GRU for improving forecasting performance. Computers & Chemical

Engineering, 155, 107513.

Cen, Z., & Wang, J. (2019). Crude oil price prediction model with long short-term memory deep

learning based on prior knowledge data transfer. Energy, 169, 160–171.

Chen, Y. C., & Huang, W. C. (2021). Constructing a stock-price forecast CNN model with gold and

crude oil indicators. Applied Soft Computing, 112, 107760.

Chen, Y., He, K., & Tso, G. K. (2017). Forecasting crude oil prices: A deep learning-based model.

Procedia computer Science, 122, 300–307.

Chi, G., Uddin, M. S., Abedin, M. Z., & Yuan, K. (2019). Hybrid model for credit risk prediction:

An application of neural network approaches. International Journal on Artiﬁcial Intelligence

Tools, 28(05), 1950017.

Datta, R. K., Sajid, S. W., Moon, M. H., & Abedin, M. Z. (2021). Foreign currency exchange rate

prediction using bidirectional long short-term memory. In The big data-driven digital economy:

Artiﬁcial and computational intelligence (pp. 213–227). Springer.

Ghaffari, A., & Zare, S. (2009). A novel algorithm for prediction of crude oil price variation based

on soft computing. Energy Economics, 31(4), 531–536.

Grange, R., & Hand, D. (1987). A review of the effects of atmospheric humidity on the growth of

horticultural crops. Journal of Horticultural Science, 62(2), 125–134.

Huang, L., & Wang, J. (2018). Global crude oil price prediction and synchronization-based

accuracy evaluation using random wavelet neural network. Energy, 151, 875–888.

Kaur, J., Nathani, N., & Kaur, M. (2013). Oil price ﬂuctuations and FOREX market: Evidence from

OPEC countries. International Journal of Research in Business and Technology, 3(3), 215–235.

Kulkarni, S., & Haidar, I. (2009). Forecasting model for crude oil price using artiﬁcial neural

networks and commodity futures prices. arXiv preprint arXiv: 0906.4838.

Lu, Q., Sun, S., Duan, H., & Wang, S. (2021). Analysis and forecasting of crude oil price based on

the variable selection-LSTM integrated model. Energy Informatics, 4(2), 1–20.

An Ensemble LGBM (Light Gradient Boosting Machine) Approach for Crude... 165

Nonejad, N. (2021). Predicting equity premium by conditioning on macroeconomic variables: A

prediction selection strategy using the price of crude oil. Finance Research Letters, 41, 101792.

Norouzi, N., & Fani, M. (2020). Black gold falls, black plague arise - An Opec crude oil price

forecast using a gray prediction model. Upstream Oil and Gas Technology, 5, 100015.

Rahman, M. M., Guotai, C., Das Gupta, A., Hossain, M., & Abedin, M. Z. (2021). Impact of early

COVID-19 pandemic on the US and European stock markets and volatility forecasting. Eco-

nomic Research-Ekonomska Istraživanja, 2021,1–18.

Ratti, R. A., & Vespignani, J. L. (2013). Why are crude oil prices high when global activity is weak?

Economics Letters, 121(1), 133–136.

Rufo, D. D., Debelee, T. G., Ibenthal, A., & Negera, W. G. (2021). Diagnosis of diabetes mellitus

using gradient boosting machine (LightGBM). Diagnostics, 11(9), 1714.

Shajalal, M., Hajek, P., & Abedin, M. Z. (2021). Product backorder prediction with deep neural

network on imbalance data. International Journal of Production Research,1–18.

Shakoor, M. T., Rahman, K., Rayta, S. N., & Chakrabarty, A. (2017, July). Agricultural production

output prediction using supervised machine learning techniques. In 2017 1st International

Conference on Next Generation Computing Applications (Next Comp) (pp. 182–187).

Vo, A. H., Nguyen, T., & Le, T. (2020). Brent oil price prediction using Bi-LSTM network.

Intelligent Automation and Soft Computing, 26(6), 1307–1317.

Wang, J., Niu, T., Du, P., & Yang, W. (2020). Ensemble probabilistic prediction approach for

modeling uncertainty in crude oil price. Applied Soft Computing, 95, 106509.

Welch, I., & Goyal, A. (2008). A comprehensive look at the empirical performance of equity

premium prediction. The Review of Financial Studies, 21(4), 1455–1508.

Wu, B., Wang, L., Lv, S. X., & Zeng, Y. R. (2021). Effective crude oil price forecasting using new

text-based and big-data-driven model. Measurement, 168, 108468.

Yang, Y., Guo, J. E., Sun, S., & Li, Y. (2021). Forecasting crude oil price with a new hybrid

approach and multi-source data. Engineering Applications of Artiﬁcial Intelligence, 101,

104217.

Zhao, Y., Li, J., & Yu, L. (2017). A deep learning ensemble approach for crude oil price forecasting.

Energy Economics, 66,9–16.

Zhao, Y., Zhang, W., Gong, X., & Wang, C. (2021). A novel method for online real-time

forecasting of crude oil price. Applied Energy, 303, 117588.

167

Model Development for Predicting

the Crude Oil Price: Comparative

Evaluation of Ensemble and Machine

Learning Methods

Mahmudul Hasan, Ushna Das, Rony Kumar Datta,

and Mohammad Zoynul Abedin

Abstract The crude oil market is unstable, and its price is highly volatile. Due to the

Covid-19 pandemic, the price of crude oils goes up and down in a short period of

time. Future plans and projects’policies depend directly and indirectly on the future

price of crude oil. So, the aim of this study is to predict the price of crude oil by using

machine learning and ensemble algorithm, as well as to show the comparison of

performance of Ada Boost, Bagging Lasso and Support Vector Regression model.

The study uses crude oil price time series data for analysis and to form a model to

predict future price. The actual vs. predicted curve is used to show the performance

of each algorithm individually. Analysis shows that the ensemble AdaBoost algo-

rithm displays better performance than other algorithms. The result is validated using

mean square error (MSE), root mean square error (RMSE), mean absolute error

(MAE), mean absolute percentage error (MAPE), two accuracy score function

variance score, and R

score. This study will help the stakeholders of the crude oil

industry in making decisions and formulating policies based on forecasted crude oil

prices.

M. Hasan

Department of Computer Science and Engineering, Hajee Mohammad Danesh Science and

Technology University, Dinajpur, Bangladesh

U. Das

Department of Electrical and Electronic Engineering, Hajee Mohammad Danesh Science and

Technology University, Dinajpur, Bangladesh

R. K. Datta

Department of Finance and Banking, Hajee Mohammad Danesh Science and Technology

University, Dinajpur, Bangladesh

e-mail: rony.datta@hstu.ac.bd

M. Z. Abedin (✉)

Department of Finance, Performance and Marketing, Teesside University International Business

School, Teesside University, Middlesbrough, Tees Valley, UK

e-mail: m.abedin@tees.ac.uk

M. Z. Abedin, P. Hajek (eds.), Novel Financial Applications of Machine Learning

and Deep Learning, International Series in Operations Research & Management

Science 336, https://doi.org/10.1007/978-3-031-18552-6_10

Keywords Crude oil · Price prediction · Ensemble learning · Machine learning

168 M. Hasan et al.

1 Introduction

Crude oil, also known as liquid petroleum, accumulates in porous rock formations in

the Earth’s crust and is used as fuels or for the processing of chemical compounds.

Crude oil is not only yellowish-black oil; it is a quarry of the golden possibilities that

form as a result of the decomposition of organic material within the crust of the

Earth. In global warming and intense impacts of environmental issues, roughly

discouraging to reduce the usage of crude oil, almost all of us rely on renewable

energy in order to save the environment and protect the future generation. But crude

oil is the most valuable energy resource in the present world. Crude oil is essential

for various chemical industrial products, including plastics, solvents, fertilizers, and

pesticides (Dhifqaui et al., 2022). The price of crude oil and the global economy are

interrelated and depend on each other. If for any reason the price of crude oil

ﬂuctuates, there will be a massive change in the activities of the global economy

(Baumeister & Kilian, 2016). The inﬂuence factors of crude oil price include supply

and demand, ﬁnance factor, and technology are directly inﬂuencing the change of the

interior and exterior environment of the crude oil market. Day by day, the inﬂuenc-

ing factors become perplexing and diverse. So, accurate crude price forecasting is a

really tough process nowadays (Hamilton, 2009; Kilian & Murphy, 2014; Zhang

et al., 2015; Wang et al., 2015; Tang et al., 2012). Many researchers have applied

various machine learning methods to predict the price of crude oil. In this way, the

Support Vector Machine and the Neural Network are generally used (Zhao et al.,

2017). In addition to the Multi-Recurrent Network (Orojo et al., 2019), LSTM

(Dhifqaui et al., 2022; Hajek & Abedin, 2020), ARIMA (Abdollahi & Ebrahimi,

2020) and the Deep Belief Network (Chen et al., 2019) have been used to predict the

price of crude oil. The high prediction accuracy of the crude oil price is beneﬁcial in

asset assignment, to mitigate risks for investors and ﬁnancial policy adjustment for

policy makers. It is working as a safeguard for national security and to naturalize the

economic growth of the country (Abedin et al., 2019; Guotai et al., 2017). Data

processing and a suitable model selection have been splayed the possibility of

obtaining a high prediction accuracy (Abedin et al., 2021). In this research, the

authors applied AdaBoost, Bagging Lasso, and Support Vector Regression machine

learning methods to predict the crude oil price with reliable. And ﬁnally, the authors

compare all three methods with various error measurement techniques and reach a

decision that AdaBoost is better than other algorithms. It is a highly suitable method

to predict the price of crude oil.

2 Related Literature

Three factors such as supply and demand, ﬁnance, and technology are inﬂuencing

the price of crude oil (Lu et al., 2021). Considering the three factors, we have to

clarify the best congruent forecast scheme among others. Time series models,

econometric models, qualitative models, and artiﬁcial intelligence models are

immensely operable for oil price forecasting and modeling. In recent era, anticipat-

ing the price of crude oil is a great blessing for many large and tiny industries,

individuals and countries (Abedin et al., 2020). To predict the price of oil, many

economists and analysts use autoregressive moving average models and vector

regression models (Chai et al., 2022). Artiﬁcial intelligence methods and traditional

econometric models are two highly responsive methods to predict the price of crude

oil at present. In the accuracy issue, artiﬁcial intelligence methods are more com-

patible than traditional econometric models (Song et al., 2020). Ensemble probabi-

listic prediction is given more efﬁciently than deterministic prediction. The

deterministic prediction contains prediction errors that create a discrepancy in

ﬁnancial decision-making in the crude oil market. But the ensemble probabilistic

method attempts to overcome all difﬁculties and mitigate all risks (Satu et al., 2020).

There is a hectic relationship between global economics and crude oil prices. For

crude oil market indices throughout the world, West Texas Intermediate Crude oil

and Brent Crude oil are the most important (Li et al., 2021). The forecasting level is

increased by a good data length. Moreover, the length of the data on a daily basis

gives a good forecasting level compared to weekly and monthly (Zhao et al., 2021).

Due to economic crises, geopolitics, and unforeseen occurrences, the price of crude

oil is immensely impacted. The model collocation inﬂuences the prediction ability of

the model. The validity of crude oil price forecasts would be affected by erroneous

model collocation (Yu et al., 2016). Linearity, non-linearity, hysteresis, structural

discontinuities, and instability are all aspects of crude oil time series. The

decomposing algorithm may be used to create sub-series or components with

linearity, non-linearity, and instability (Yu et al., 2016). In the crude oil price, to

detect the latent nonlinear features, traditional methods may not be feasible. As a

result, a new technique is required to overcome the drawbacks of conventional

methods. According to prior studies, artiﬁcial intelligence models with robust self-

learning capabilities, such as support vector machines (SVMs), artiﬁcial neural

networks (ANNs), and other intelligence algorithms, have become increasingly

popular for crude oil price predictions. Empirical evidence indicates that they

outperformed traditional methods. AI models admit its radical limitations such as

time consuming, slack convergence, and local minima (Yang et al., 2021). For

analyzing tangle and anomaly data, the “decomposition and ensembled”principle

is deliberated as an excellent tool (Datta et al., 2021). Data preparation, which

includes data cleaning, data transformation, and data reduction, is a critical stage

whose main purpose is to generate ﬁnal data sets that are appropriate and precise for

future predictions. In the forecasting literature, there are a variety of strategies for

data reduction, including feature selection and future extraction. Feature selection

can detect and eliminate as many redundant and unnecessary characteristics as

possible. Most crude price forecasting research employs feature selection for data

reduction because features maintain their original characteristics, allowing for

improved model interpretation. Feature selection only keeps valid variables by

deﬁning a threshold, so discarding a lot of important data, whereas feature extraction

reduces the original feature space to a simpler one, retaining more data (Abedin

et al., 2019).

Model Development for Predicting the Crude Oil Price:... 169

170 M. Hasan et al.

3 Methodology

To predict the price of crude oil, the traditional machine learning analysis technique

is applied. Before splitting the data into training and testing, it is preprocessed. For

training and testing data, we divided the dataset into an 80:20 ratio at random.

Machine learning and ensemble techniques are used to build the analytical model,

which is then trained using the training data to provide projected values using the

testing values as input. Figure 1shows a block diagram of our proposed

methodology.

3.1 Dataset

The dataset that is used for the analysis is the price of the Brent crude oil –Europe

data. It is taken from the US Energy Information Administration. It releases as spot

prices and its price is in Dollars per barrel. Data frequency is daily, but not seasonally

adjusted. It is time series data from May 20, 1987 to September 10, 2021, and the

total number of observations is 8954. Figure 2represents the information about the

dataset.

The price of crude oil was stable during the period 1987 to 2000. After this time,

the price increased. During 2008–2009 it was maximum, and then the price went

down. Between 2011 and 2015, the price was in a stable situation and after this

period it started falling. In 2020 the price of crude oil fell due to the Covid-19

pandemic. The situation is going to go well now and the price is also increasing. The

plot clearly indicates that there is a great impact of Covid-19 on the price of crude oil.

The above discussion indicates that the market for crude oil is not fully stable. Many

variables are responsible for varying this price. The prediction of this market is really

difﬁcult and needs special and deep analysis. The numerical description of the

dataset is given in Table 1.

Fig. 1 Block diagram of the proposed methodology for predicting the price of crude oil

Model Development for Predicting the Crude Oil Price:... 171

100

120

140

160

20-05-1987

20-05-1989

20-05-1991

20-05-1993

20-05-1995

20-05-1997

20-05-1999

20-05-2001

20-05-2003

20-05-2005

20-05-2007

20-05-2009

20-05-2011

20-05-2013

20-05-2015

20-05-2017

20-05-2019

20-05-2021

OIL PRICE

DATE

Crude Oil Price Data From 1987 to 2021

Fig. 2 Crude oil price data from 1987 to 2021

Table 1 Statistical descrip-

tion of the Brent crude oil data Mean Standard Deviation Min Max

46.75337 32.01776 9.10000 143.95000

3.2 Description of the Algorithms

AdaBoosting

Boosting is a kind of ensemble technique that improves prediction accuracy by

converting a number of weak learners into strong learners. The Boosting algorithm

works on the principle that the ﬁrst model is developed in the training data set and the

second model is constructed to correct the ﬁrst model errors. This procedure is

iterated until the errors are minimized and the data instances are accurately predicted.

For each feature, this algorithm generates a weak regressor. Because the weight of

effectively calculated samples will be suitably lowered, while the weight of

misclassiﬁed samples will be appropriately raised, the original classiﬁer does not

require a high accuracy if somehow the accuracy is higher than that of random. As a

result, the sample distribution is altered. A strong regressor with improved perfor-

mance may be created by merging the weak samples acquired from each cycle. The

features that these powerful classiﬁers employ are well-classiﬁed (Fig. 3).

Bagging Lasso

The lasso is a shrinking approach similar to the ridge, but with some key distinctions.

The lasso regression cost function may be deﬁned as follows:

172 M. Hasan et al.

Fig. 3 Block diagram of the operation of the AdaBoost algorithm

i=1

yi-b

ðÞ

2=X

i=1

yi-X

j=0

wj×xij

þλX

j=0

for some t>0, X

j=0

wj

<t:

ð1Þ

The key distinction between the formulations of the cost function of the ridge and

lasso regression is that in the lasso regression, instead of calculating the square of the

coefﬁcients, the magnitudes are factored into the equation. This method of regular-

ization (L1) might result in a zero coefﬁcient, which means that some characteristics

are completely ignored when evaluating the output. As a result, lasso regression not

only aids in the reduction of over-ﬁtting but also in the selection of characteristics

that make the model easier to understand.

Bagging Lasso is an ensemble algorithm constructed by the bagging ensemble

procedure, where Lasso is used as a base algorithm. The data is bagged into different

parts and then trained by the Lasso regression. Finally, the ﬁnal results emerge and

give better accuracy than the base Lasso model.

SVR (Linear, RBF, Polynomial)

In today’sworld, the most widely utilized and high-performance algorithm is the

support vector machine. This is a supervised machine learning approach that may be

used to classify and predict data. However, the authors can employ this learning

approach to solve regression problems. The goal of SVM is to build a model (based

on the training data). Given only the test data features, the model anticipates the

output of the target values of the test data. Linear SVM and Kernel SVM are the two

forms of SVM that are currently accessible. Linear SVM is an incredibly fast

machine learning approach for solving multiclass problems from large datasets

(Fig. 4).

Model Development for Predicting the Crude Oil Price:... 173

Fig. 4 Block diagram of Support Vector Machines

SVM implements an exclusive proprietary version of a linear support vector

machine design algorithm. This algorithm classiﬁes the data by generating a decision

boundary based on the support vector point (Yang et al., 2021). In some instances,

the accuracy of SVM is higher than that of other classiﬁcation algorithms. Kernel

SVM is employed for nonlinear data categorization because the data in the real world

is not as straightforward as the data in the previous picture. The Kernel SVM is a

modiﬁed SVM algorithm that may be used to categorize this type of data. SVM’s

kernel contains a number of arithmetic operations. The functions take data as input

and transform them into the format necessary. There are various types of mathemat-

ical function. Polynomial, sigmoid, linear, nonlinear, and radial basis functions, for

example.

3.3 Performance Measures

MAE: It is nothing more than an arithmetic average of the absolute errors. It is the

simplest measurement for computing forecast accuracy. It measures the accuracy for

a continuous variable as follows:

MAE =1

i=1

yi -y

:ð2Þ

MSE: The Mean Square Error is narrated as an average of the difference between

actual and estimated value. In this procedure, all errors are positive. It is highly

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

sensitive to outliers. The small value of this model represents a better model. The

MSE is deﬁned as follows:

174 M. Hasan et al.

MSE =1

nXn

i=1y-yiðÞ

:ð3Þ

RMSE: The Root Mean Square Error is the root of the mean of the square of all of

the errors. It is a standard way to measure the error of a model given as:

RMSE =1

i=1

y-yiðÞ

:ð4Þ

MAPE: The average percentage error for each time period minus genuine values

divided by genuine values is how MAPE determines this reliability as a percentage:

MAPE =1

nXn

i=1

y-yiðÞ

y:ð5Þ

Variance Score: The variance is a metric to determine how variable something

is. To calculate it, the mean square deviation is used. The dispersion of data collected

is measured by variation. The greater the difference in average, the greater the data

spread.

Score:R

varies from 0 to 1. It determines how well the data match the

regression line. For predictive models, a low R

value is usually a poor indication.

An excellent model may display a little value in some circumstances.

4 Results and Discussion

In this paper, two ensemble algorithms are used to predict the crude oil price. The

actual vs. predicted values of the algorithms are given in Fig. 5.

Figure 5indicates the actual vs. predicted curve of crude oil using AdaBoost. The

red color indicates the actual values and the blue color indicates the predicted values

of the Brent oil price data. The curve shows that the performance of lass is good and

after evaluation we get 0.00932 MAE, 0.00015 MSE, 0.01235 RMSE, and 0.24785

MAPE error, which are tabulated in Table 2. From Table 3, we see that the variance

score is 0.98 and the R

score is 0.98. This curve also indicates that the price of crude

oil does not have any speciﬁc rules. It can fall at any time and increase at any time.

The result of the remaining algorithms is shown and discussed below, one by one.

Figures 6,7,8,and 9show the actual vs. predicted curve of crude oil using

Bagging Lasso Regression, SVR (Linear Kernel) Regression, SVR (RBF Kernel)

Regression, and SVR (Polynomial Kernel) Regression, respectively. The results

show that the Bagging Lasso Regression performed best in terms of MAPE error

(0.40649), while the SVR (Polynomial Kernel) Regression model was superior with

respect to MAE (0.01663), RMSE (0.01986), and R

score (0.96). In fact, the

Bagging Lasso Regression overestimates the prices of crude oil, whereas for the

SVR models, it is rather the opposite.

Model Development for Predicting the Crude Oil Price:... 175

Crude Oil Price Prediction Using Ada Boost

Crude Oil Price

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0250 500 750 1000 1250 1500 1750

Data Sample According to Time

Actual

Predicted

Fig. 5 Curve of actual vs. predicted crude oil price using AdaBoost

Table 2 Performance measurement of different algorithms for the prediction of crude oil price

Method MAE MSE RMSE MAPE

Ada Boost 0.00932 0.00015 0.01235 0.24785

Bagging Lasso Regression 0.01738 0.00046 0.02152 0.40649

Support Vector Machine (Linear) 0.01743 0.00042 0.02059 0.45646

Support Vector Machine (RBF) 0.01663 0.00039 0.01986 0.46593

Support Vector Machine (Polynomial) 0.02653 0.00098 0.03131 0.59497

Note: The best-performing algorithm is in bold

Table 3 Different types of scores of algorithms for the prediction of crude oil price

Method Variance Score R

Score

AdaBoost 0.98 0.98

Bagging Lasso Regression 0.95 0.95

Support Vector Machine (Linear) 0.95 0.95

Support Vector Machine (RBF) 0.96 0.96

Support Vector Machine (Polynomial) 0.90 0.90

Note: the best-performing algorithm is in bold

Table 2represents the MAE, MSE, RMSE and MAPE error values of AdaBoost,

Bagging Lasso Regression, and different kernel functions of Support Vector Regres-

sion. It clearly indicates that all kinds of error in AdaBoost are less than those of the

other algorithms. It means that the prediction of AdaBoost is better than that of the

other algorithms. For clear understanding, we represent the errors in a line chart in

Fig. 10.

176 M. Hasan et al.

Crude Oil Price Prediction Using Bagging Lasso Regression

Crude Oil Price

Data Sample According to Time

Actual

Predicted

0.5

0.4

0.3

0.2

0.1

0.0

0 250 500 750 1000 1250 1500 1750

Fig. 6 Curve of actual vs. predicted crude oil price using Bagging Lasso Regression

Crude Oil Price Prediction Using SVM (Linear)

Crude Oil Price

Data Sample According to Time

Actual

Predicted

0.5

0.4

0.3

0.2

0.1

0.0

0250 500 750 1000 1250 1500 1750

Fig. 7 Actual vs. predicted curve of crude oil using SVR (Linear Kernel) Regression

Figure 10 represents MAE, MSE, RMSE and MAPE of three models. The orange

color represents the errors of the AdaBoost model, the yellow color represents the

errors of Bagging Lasso, the green color represents the linear SVM, the purple color

represents the RBF SVR, and the coffee color represents the Polynomial SVR. The

numeric values 1, 2, 3, 4 represent MAE, MSE, RMSE, and MAPE consecutively.

The ﬁgure clearly indicates that the error rate of AdaBoost is less than that of

Bagging Lasso.

Model Development for Predicting the Crude Oil Price:... 177

Crude Oil Price Prediction Using SVM (RBF)

Crude Oil Price

Data Sample According to Time

Actual

Predicted

0.5

0.4

0.3

0.2

0.1

0.0

0 250 500 750 1000 1250 1500 1750

Fig. 8 Curve of actual vs. predicted crude oil price using SVR (RBF Kernel) Regression

Crude Oil Price Prediction Using SVM (Polynomial)

Crude Oil Price

Data Sample According to Time

Actual

Predicted

0.5

0.4

0.3

0.2

0.1

0.0

0 250 500 750 1000 1250 1500 1750

Fig. 9 Curve of actual vs. predicted crude oil price using SVR (Polynomial Kernel) Regression

Table 3shows the Variance and R

scores for the compared methods, suggesting

that AdaBoost also outperforms the Bagging Lasso Regression and the three Support

Vector Regression Model in terms of explained variance, which conﬁrms that the

predicted values obtained by AdaBoost ﬁt well the actual oil prices.

178 M. Hasan et al.

Fig. 10 Comparison of different methods

5 Conclusion and Future Work

The purpose of this study is to forecast the price of Brent crude oil. For analysis, we

use the SVR machine learning algorithm and two ensemble techniques Ada Boost

and Bagging Lasso Regression. The AdaBoost ensemble machine learning tech-

nique outperforms others in terms of overall performance. All the data are presented

in tabular and graphical format. The performance of the other algorithms is equally

satisfactory, and the error rates are too low. This study helps everyone involved in

this industry make difﬁcult decisions that are directly or indirectly inﬂuenced by

crude oil prices.

In the future, direct and indirect factors can be included, and deep neural network

can be used for better prediction. In addition, a website can be developed based on

the analysis that can show real-time analysis on the future price of crude oil data.

References

Abdollahi, H., & Ebrahimi, S. B. (2020). A new hybrid model for forecasting Brent crude oil price.

Energy, 200, 117520.

Abedin, M. Z., Guotai, C., Moula, F. E., Azad, A. S., & Khan, M. S. U. (2019). Topological

applications of multilayer perceptrons and support vector machines in ﬁnancial decision support

systems. International Journal of Finance & Economics, 24(1), 474–507.

Abedin, M. Z., Chi, G., Uddin, M. M., Satu, M. S., Khan, M. I., & Hajek, P. (2020). Tax default

prediction using feature transformation-based machine learning. IEEE Access, 9, 19864–19881.

Abedin, M. Z., Moon, M. H., Hassan, M. K., & Hajek, P. (2021). Deep learning-based exchange

rate prediction during the COVID-19 pandemic. Annals of Operations Research, 2021,1–52.

https://doi.org/10.1007/s10479-021-04420-6

Model Development for Predicting the Crude Oil Price:... 179

Baumeister, C., & Kilian, L. (2016). Forty years of oil price ﬂuctuations: Why the price of oil may

still surprise us. Journal of Economic Perspectives, 30(1), 139–160.

Chai, S., Chu, W., Zhang, Z., & Abedin, M. Z. (2022). Dynamic nonlinear connectedness between

the green economy, clean energy, and stock price: The impact of the COVID-19 pandemic.

Annals of Operations Research, 2022,1.

Chen, J. H., Hao, Y. H., Wang, H., Wang, T., & Zheng, D. W. (2019). Futures price prediction

modeling and decision-making based on DBN deep learning. Intelligent Data Analysis, 23(S1),

53–65.

Datta, R. K., Sajid, S. W., Moon, M. H., & Abedin, M. Z. (2021). Foreign currency exchange rate

prediction using bidirectional long short-term memory. In A. M. A. Musleh Al-Sartawi (Ed.),

The big data-driven digital economy: Artiﬁcial and computational intelligence. Studies in

computational intelligence (Vol. 974). Springer. https://doi.org/10.1007/978-3-030-73057-4_

Dhifqaui, Z., Khalfqoui, R., Abedin, M. Z., & Shi, B. (2022). Quantifying information transfer

among clean energy, carbon, oil, and precious metals: A novel transfer entropy-based approach.

Finance Research Letters, 49, 103138.

Guotai, C., Abedin, M. Z., & Moula, F. E. (2017). Modeling credit approval data with neural

networks: An experimental investigation and optimization. Journal of Business Economics and

Management, 18(2), 224–240.

Hajek, P., & Abedin, M. Z. (2020). A proﬁt function-maximizing inventory backorder prediction

system using big data analytics. IEEE Access, 8, 58982–58994.

Hamilton, J. D. (2009). Understanding crude oil prices. The Energy Journal, 30(2), 179–206.

Kilian, L., & Murphy, D. P. (2014). The role of inventories and speculative trading in the global

market for crude oil. Journal of Applied Econometrics, 29(3), 454–478.

Li, R., Hu, Y., Heng, J., & Chen, X. (2021). A novel multiscale forecasting model for crude oil price

time series. Technological Forecasting and Social Change, 173, 121181.

Lu, Q., Sun, S., Duan, H., & Wang, S. (2021). Analysis and forecasting of crude oil price based on

the variable selection-LSTM integrated model. Energy Informatics, 4(2), 1–20.

Orojo, O., Tepper, J., McGinnity, T.M., & Mahmud, M. (2019, December). A multi-recurrent

network for crude oil price prediction. In 2019 IEEE Symposium Series on Computational

Intelligence (SSCI) (pp. 2940–2945). IEEE.

Satu, M. S., Rahman, S., Khan, M. I., Abedin, M. Z., Mahmud, M., & Kaiser, M. S. (2020,

September 19). Towards improved detection of cognitive performance using bidirectional

multilayer long short-term memory neural network. In The proceedings of LNAI: The 13th

international conference on brain informatics: Virtual conference, pp. 297–306.

Song, X., Liu, Y., Xue, L., Wang, J., Zhang, J., Wang, J., & Cheng, Z. (2020). Time-series well

performance prediction based on Long Short-Term Memory (LSTM) neural network model.

Journal of Petroleum Science and Engineering, 186, 106682.

Tang, L., Yu, L., Wang, S., Li, J., & Wang, S. (2012). A novel hybrid ensemble learning paradigm

for nuclear energy consumption forecasting. Applied Energy, 93, 432–443.

Wang, Y., Wu, C., & Yang, L. (2015). Forecasting the real prices of crude oil: A dynamic model

averaging approach. Available at SSRN 2590195.

Yang, Y., Guo, J., Sun, S., & Li, Y. (2021). Forecasting crude oil price with a new hybrid approach

and multi-source data. Engineering Applications of Artiﬁcial Intelligence, 101, 104217.

Yu, L., Dai, W., & Tang, L. (2016). A novel decomposition ensemble model with extended extreme

learning machine for crude oil price forecasting. Engineering Applications of Artiﬁcial Intelli-

gence, 47, 110–121.

Zhang, J. L., Zhang, Y. J., & Zhang, L. (2015). A novel hybrid method for crude oil price

forecasting. Energy Economics, 49, 649–659.

Zhao, Y., Li, J., & Yu, L. (2017). A deep learning ensemble approach for crude oil price forecasting.

Energy Economics, 66,9–16.

Zhao, Y., Zhang, W., Gong, X., & Wang, C. (2021). A novel method for online real-time

forecasting of crude oil price. Applied Energy, 303, 117588.

Part IV

Emerging Technologies in Financial

Education and Healthcare

183

Discovering the Role of M-Learning Among

Finance Students: The Future of Online

Education

Armana Hakim Nadi, Syed Far Abid Hossain, Al Mahmud Hasan,

Mahbuba Rahman Soﬁn, Saadman Shabab, Md. Ahmedul Islam Sohan,

and Chunyun Yuan

Abstract The chapter aims to explore the role of m-learning among ﬁnance stu-

dents with an additional focus on the future of online higher education. The key

reason to conduct the study is to explore the hidden issues of m-learning for the

students majoring in ﬁnance, especially in the online classroom setting. The study

used a qualitative research approach to discover the phenomenon. The authors

conducted a thorough literature review of the existing literature and attempted to

fulﬁll the research gap following the qualitative research approach. The result shows

that digitalized education provides the opportunity for ﬁnance major students to

access ﬁnancial markets using the Internet and gain personal and professional

knowledge in a better way rather than traditional learning. The result also discovers

a signiﬁcant positive relationship between m-learning and online educational effec-

tiveness. Only the students of Finance were the participants which may affect the

generalizability. The study presents signiﬁcant implications for education

policymakers and practitioners. The study ﬁlls the gap in the current literature by

discovering the role of m-learning in the online educational setting for ﬁnance major

students.

A. H. Nadi

Bangladesh University of Professionals, Dhaka, Bangladesh

S. F. A. Hossain (✉)

BRAC Business School, BRAC University, Dhaka, Bangladesh

A. M. Hasan · M. R. Soﬁn · M. A. I. Sohan

IUBAT University, Dhaka, Bangladesh

e-mail: asohan@iubat.edu

S. Shabab

North South University, Dhaka, Bangladesh

C. Yuan

The School of Economic Management, University of Arts & Sciences, Baoji, China

M. Z. Abedin, P. Hajek (eds.), Novel Financial Applications of Machine Learning

and Deep Learning, International Series in Operations Research & Management

Science 336, https://doi.org/10.1007/978-3-031-18552-6_11

Keywords m-learning · Finance students · Higher education · Online education ·

Traditional learning

184 A. H. Nadi et al.

1 Introduction

Owing to technological advancements in the education sector, the signiﬁcance of

Mobile Learning has skyrocketed to a great extent in the modern world. The impact

of m-learning on education is a global phenomenon today. Especially in this post-

COVID era, m-learning has had an immense impact on scholars and students from

around the globe. After the outbreak of the COVID pandemic around the world,

education through “traditional learning”that involves direct teaching or an in-person

approach in classrooms has become near impossible. In contrast, “e-learning”,

although proven to be somewhat useful in conducting online sessions, lacks effective

retention of the material studied by the students. Smartphones are increasingly

serving as the primary source of entertainment, information, communication, and

other resources during times of crisis when individuals must stay at home for a

longer period of time (Khan et al., 2022). Smartphones are becoming more and more

the primary platform for m-learning. M-learning is complementary to both tradi-

tional learning and e-learning (Kumar Basak et al., 2018). In recent years, big data

science, biomedical computing, and precision medicine have all beneﬁted greatly

from the steadily increasing desire to introduce algorithms in machine learning in

conjunction with multi-omic data for detecting potential genotype–phenotype links

(Khan et al., 2021). M-learning has combined the best of both worlds and introduced

mobility in education, thus paving the way for portable education in the process.

Imagine the following scenario: You have some newly recruited employees

whom you have been training for a month. At the end of their training, they should

be able to perform all the tasks assigned to them when they join the workplace.

However, this is highly unlikely. “Research shows that in one hour people will have

forgotten an average of 50 percent of the information you presented. Within

24 hours, they have forgotten an average of 70 percent of new information, and

within a week they forget to claim an average of 90 percent of it.”(“Brain science:

The forgetting curve–the dirty secret of corporate training”,2019). So, when your

new employees are ready to start working, they would be lucky if they remembered

at least half of what they had learned during training, let alone the name of the

trainer. However, this situation can be improved through m-learning to some extent.

M-learning would help these employees in the hands-on situation with the resources

and training they need at that time. Employees can simply pull out their cell phones

to get just-in-time training or supporting materials that would allow them to identify

and perform the task at hand. Since employees learn the task at the very moment they

perform it, they will be able to apply the majority of what they have just learned to

the task at hand. Thus, M-learning educates learners by providing bite-sized infor-

mation, the on-the-go, and just-in-time to perform a task or solve a problem.

Access to ﬁnancial technology services is relatively well-known to people who

are inﬂuenced by knowledge of ﬁnancial services (Hasan et al., 2022). Financial

behavior has improved signiﬁcantly through ﬁnancial literacy (Wahyudi et al.,

2020). However, to achieve ﬁnancial literacy, it is imperative to implement

m-learning in the present world scenario.

Discovering the Role of M-Learning Among Finance Students: The Future... 185

In the world of higher education, speciﬁcally in the Finance major, scholars are

gradually implementing M-learning. It is well known that higher education in

Financial Management is heavily focused on mathematical data calculation and

analysis. Hence, the traditional education approach in this aspect demands ﬁnance

students to memorize a myriad of formulae. However, m-learning signiﬁcantly

reduces the inconvenience for students and contributes to efﬁcient learning. The

current literature has explored IoT (Internet of Things) in education (Nguyen et al.,

2022) with an opportunity to 6G in the near future; class imbalanced prediction

(Abedin et al., 2022); deep learning in the contemporary era (Abedin et al., 2021);

technological applications (Abedin et al., 2019); the way to achieve education

sustainability with advanced technology usage (Hossain et al., 2022) and TPACK

development with smartphone usage (Hossain et al., 2021).

In this chapter, we will study the role m-learning plays in students who are

majoring in ﬁnance in their higher education.

2 Literature Review on Mobile Technologies in Teaching

M-learning or Mobile learning has become one of the most talked about topics in the

world right now. And due to recent events, m-learning has become a crucial part of

daily life for everyone. Starting from teachers to corporate employees, everyone has

adopted m-learning as a natural part of their learning routine to stay up-to-date even

in this pandemic. Mobile learning, simply put, is any form of education where the

central technology used for learning is a handheld or palmtop device. Devices such

as smartphones, tablets, and even laptops can be used for m-learning (Guy, 2009).

However, there is a common misconception that using personal desktop computers

for educational purposes falls under m-learning. This is clearly wrong since

m-learning takes advantage of the mobility of handhelds and to provide the user

with the opportunity to learn anywhere and at any time. The high success rate of

m-learning can mainly be attributed to the high penetration of mobile phones around

the world. Even in 2015 studies had shown that mobile phones successfully pene-

trated 97% of the world (Criollo-C et al., 2018). In recent times, there are almost

8 billion mobile devices in the world. This only goes to show the popularity that

mobile devices have gained over the years. In addition to the obvious, m-learning

has many advantages. The ﬁrst obvious advantage, which is evident from the name,

is mobility. M-learning has allowed users to learn and teach even when they are

away from their homes, schools, ofﬁces, and any other conventional learning

locations. Another important advantage of M-learning is accessibility. Gone are

the days when users would need a full desktop computer to access the Internet.

Instead, smartphones give users the ability to access any content on the Internet in

the palm of their hand. To add to this, smartphones now have added functionality

that desktops cannot provide, like gyro sensors that can be used to view 3D images.

Mobile learning helps both students and teachers. Mobile learning enables students

to self-regulate their learning and also allows teachers to customize instructions as

they see ﬁt (Naciri et al., 2020). Mobile learning is unique because it does not bind

students to a certain place or a certain time frame. Students can access the learning

materials anytime and anywhere which introduces the idea of training at their

convenience (Bazhenov, 2011).Another study shows that university students have

a greater ability to learn a foreign language when they do so using a smartphone.

This research also shows that although M-learning may not be able to completely

replace traditional learning, it can, however, be used to complement to achieve

enhanced teaching outcomes (Klimova, 2019). It must also be kept in mind that

factors such as information quality and information quality also have an impact on

how likely students are to and are satisﬁed with mobile learning (Almaiah &

Alismaiel, 2018). Other factors such as the ability of a teacher to make use of

m-learning without sacriﬁcing the quality of education in a traditional or physical

classroom can also play a signiﬁcant role in the adoption of mobile learning by the

student (Pedro et al., 2018).

186 A. H. Nadi et al.

3 The Impact of M-Learning on Finance Students

In order to fully realize the potential and the impact of m-learning among Finance

students, we must ﬁrst understand how a Finance student can utilize m-learning. In a

general sense, there are three ways mobile learning can be used: educational video

content on video portals, mobile apps providing bite-sized lessons on topics, and

ﬁnally, group study through learning groups on social media. It goes without saying

that there is a smartphone in almost every pocket in the world at present. If you own a

smartphone, it is safe to say that you are familiar with video portals such as

YouTube, Vimeo, Daily Motion, etc. These video portals offer numerous tutorial

videos on thousands of topics. A ﬁnance student can also easily ﬁnd tutorials on

different topics such as ﬁnancial ratio calculation, wealth management, corporate

ﬁnance, investment banking, and many more. It is only a matter of searching for a

speciﬁc topic and watching a video.

Mobile apps have always been and will continue to be an integral part of

smartphones. The variety of apps is endless, to say the least. Engagement with

educational apps improves the students’competencies (Camilleri & Camilleri,

2019). Many educational apps prove to be useful to Finance students in their higher

education. For example, Android apps in Google Play Store such as “Finance

Formulas”and “Financial Ratio Calculator”help students learn and implement a

myriad of formulas required in ﬁnance education. Social media also play a crucial

role in m-learning for ﬁnance students. Social media are argued to have the potential

to bridge formal and informal learning via a digital culture of participation

(Greenhow & Lewin, 2019). Social media can be used for educational purposes in

several ways such as enhancing communication and interaction between students

and between teachers and students, as well as promoting student engagement as it

allows intimidated, shy, or bored students to share thoughts and express his or her

opinion more comfortably (Faizi et al., 2013). Furthermore, educational groups and

pages on social media platforms such as Facebook facilitate ﬁnance students to

discuss and perform group studies on various topics of interest. Students with

Finance major in higher education can also stay up-to-date on innovations in their

ﬁeld through social media. Last but not least, social media provides networking

opportunities to Finance students with successful individuals in their career paths.

Discovering the Role of M-Learning Among Finance Students: The Future... 187

4 Available Mobile Applications for Online Platforms

Mobile devices and applications to support teaching and learning (m-learning) have

received attention in education. In many nations and regions, the spread of Covid-19

has resulted in a rapid shift from traditional to online education platforms. The use of

technology in education signiﬁcantly impacts learning, with universities serving as

the primary providers of online education (Aljaaidi et al., 2020). There are different

operating platforms, such as Android, iOS and Windows Mobile, that build mobile

apps (Hamilton, 2019). Mobile applications make educational information more

accessible, and each app has its own set of characteristics that allow it to provide

its own set of services. The mobile application also offers online educational services

through e-Books, e-Library, informative videos, and games (Jaber et al., 2021). The

use of virtual reality (AR) in education has several advantages, including improved

engagement and interaction, and can mitigate the negative consequences of face-to-

face education disruption (Criollo-C et al., 2021). Learners can access the material

anywhere and anytime with learning approaches, with just the touch of a button on

the mobile application (Baharum et al., 2020). Therefore, the mobile application is of

great beneﬁt to the learner. Mobile-Based Assessment has been increasingly popular

in higher education worldwide in recent years; even every learning material is

available through a mobile application on the mobile device (Singh et al., 2021).

However, the application of technology improves the ability of instructors to reduce

digital gaps, improve digital creativity, raise awareness, improve critical thinking,

and build reliability on the online platform (Dorouka et al., 2020). According to a

study, teachers used Live Video Streaming on numerous social media platforms

(such as live social media or linked live) to deliver online instruction to increase

student engagement (Chen et al., 2021). Furthermore, mobile learning technologies

provide web-based teaching and learning platforms for teachers and learners around

the world (Akour et al., 2021). M-learning technology assists teachers in saving time

by allowing them to check assignments completed by students, solve numerical

methods from the section of calculus for higher mathematics, and use a QR code

application to determine whether it is correct or incorrect (Zhylenko et al., 2020).

Based on research, students’learning activities and motivation improve after

adopting an English game-based Mobile Application (EBMA) in learning (Soﬁana

& Mubarok, 2020). The revolution technology provides numerous applications

available for online learning. Currently, renowned Google Drive applications (docs,

spreadsheets, presentations, forms) are gaining popularity and may be utilized

efﬁciently in online education to facilitate communication between academic pro-

fessionals and students. The learning process, Google Keep, Microsoft Forms, and

mural.co designed to construct group work (Llerena-Izquierdo et al., 2020). Day by

day, many free online application resources are being updated and new features are

also being added for online education. Even the availability of online learning

platforms helps students gain different skills, learning activities, and building inter-

est in learning through application. Especially during the COVID-19 epidemic,

mobile learning helped students ﬁll in the gaps in their studies (Biswas et al.,

2020). There is a great deal of interest in the use of mobile devices and technologies

for learning purposes for learners and the need to integrate them more deeply into

teacher education in all technological advancements (Connolly et al., 2021).

188 A. H. Nadi et al.

5 Online Platforms for University Students

Although institutions are introducing new areas of study to use the online learning

platform, it provides university students with more options to learn. Previously,

e-learning, distant education, and correspondence courses were commonly accepted

as non-formal education components. However, if current trends continue, it appears

that they will gradually supplant the traditional schooling system (Mishra et al.,

2020). With more and more university students wanting to study online, online

education has become a vital component of modern higher education (Australian

Government, 2011). Ted-Ed, Coursera, Google Classroom, Bakpax, Pronto, and

Skillshare are examples of some of the most popular online networking sites that will

alter the direction and route of the entire educational system in post-COVID-19

scenarios around the world (Mishra et al., 2020). Because online learning will soon

become the norm, the government, telecommunication companies, and universities

should fund the establishment of technological infrastructure throughout the country

(Chung et al., 2020). Furthermore, if students’experiences meet their expectations,

they are more likely to feel at ease and continue their studies, and likewise. If

students miss classes, want to avoid being absent while ﬁlling knowledge gaps,

they can attend online training sessions and pass the relevant online tests. The system

automatically reports test results to teachers, and when the student is successful, the

session evaluation is approved and the student is successfully assessed. To prepare

students for the ﬂuctuations of the employment market caused by machine learning

and automation, higher education must change and grow quickly and continuously.

The communicative online platform system may be linked to a university’s student

information system, allowing it to modify outreach based on students’actual pro-

gress on each required transformation activity. The design of the electronic learning

platform, on the other hand, boosts the intellectual and creative qualities of higher

education students to help them grow in their careers (Chansanam et al., 2021).

Discovering the Role of M-Learning Among Finance Students: The Future... 189

6 The Effect of Implementing M-Learning in Education

The revolution in teaching methods expands the possibilities for online education

and enhances learners’opportunities through implementing m-learning in education.

Individual acceptance of m-learning is crucial for developing countries to extend

m-learning successfully (Pratama, 2020). In addition, creating ideal circumstances

for women, middle school students, and children in rural areas to use m-learning is

critical to education. Implementing online resources is the essential factor for

learning (Herrador-Alcaide et al., 2020). Although the acceptance of m-learning in

education is effective, proper implementation is a more crucial aspect of learning.

The implementation of m-learning in education creates new approaches and educa-

tional environments based on the ﬂexible interaction between distance users connect,

anonymously or perfectly proﬁled, and between student-based communities,

allowing distance communication between students and teachers; and also between

students and machines (Fombona et al., 2020). According to the research, the

analysis found that the effect of mobile learning on student learning performance

did not vary depending on their educational level or implementation period; how-

ever, it did change depending on the course/subject (Talan, 2020). The widespread

use of mobile devices in education, as well as the popularity of transferable courses,

has resulted in many beneﬁts in terms of the learning process and outcomes, but it

has also resulted in several issues. When looking at these issues in general, they may

be classiﬁed into the following categories: technology-related hardware and software

issues, internet and infrastructure issues, mobile device screen, keyboard, and battery

issues (Kacetl & Klímová,2019). Regarding the education study, the authors explore

that informal learning contexts are most frequent in m-learning education, followed

by formal contexts and both (Aaron & Lipton, 2017). The potential for effective

teaching and learning is growing due to the implementation of m-learning in

education (Abidin & Tho, 2018). To maximize the effectiveness of M-learning,

attention should be paid to designing suitable courses to save time and improve

learning efﬁciency, increase student mobility, and offer the ﬂexibility of the course

system to learners through a variety of channels (Trinh et al., 2021). Simultaneously,

thousands of apps are available today that are challenging and problematic for both

teachers and educators (Papadakis & Kalogiannakis, 2017). Although m-learning

has been implemented at a very early stage in other countries around the world,

m-learning still solves the problems. Moreover, m-learning implementation brings

advantages to education, but it is also vital to expose learners to the convenience of

courses through various methods. One of the recent studies shows that the imple-

mentation of m-learning positively enhances students’enthusiasm for learning

kinematics as well as their self-conﬁdence (Laurens Arredondo & Valdés Riquelme,

2021). While the revolutionary adaptation of m-learning can be noticed in education

or other sectors, its implementation takes time in some nations. Based on research,

students’attitudes toward utilizing m-learning and their behavioral intents positively

inﬂuence their long-term viability in higher education (Al-Rahmi et al., 2021).

190 A. H. Nadi et al.

7 A Projection of the Available Digital Online Contents

in the Future

In the modern era, the contributions of digital online content to ﬁnancial education

are increasing. At present, there are comprehensive digital libraries that allow

students to dive even further into vital ﬁnancial topics. Websites such as

teachbanzai.com,oecd.org, and everﬁ.com provide ﬁnance students with digital

delivery of courses that include important topics in ﬁnancial management. This

digital education motivates ﬁnance students to engage themselves in an in-depth

discussion through a combination of face-to-face interaction along with online

learning. In the future, there is a huge scope for development in this sector by

generating up-to-date and innovative ideas in the ﬁeld of ﬁnancial management.

Provisions can be made for on-the-go lessons on ﬁnancial software used by organi-

zations around the world. Developments can be made by sharing lessons not only

through online media but also through interactive sessions that will allow users to

acquire ﬁrst-hand experience of the use of the ﬁnancial software.

Furthermore, with the rapid growth of technological advancements in the modern

world, it can be said without a doubt that Finance education will be heavily impacted

by innovative technologies in the future. New and improved technologies can

positively contribute to both direct education and m-learning. For example, Virtual

Reality (VR) is believed to play a crucial role in the transformation of learning and

teaching in higher education. New developments and complete immersion in the

virtual environment will undoubtedly increase students’attention (Slavova & Mu,

2018). Mark Zuckerberg announced that Facebook would change its name to Meta,

reﬂecting the new focus on creating a metaverse: a vast and integrated online world

that would cover the entire digital society and economy (Oremus, 2021). If the

possibility of this situation or in other words, the virtual universe becomes a reality,

the implementations can only be imagined as limitless. Students around the world

can gather at the same place to attend a virtual classroom session, as well as utilize

virtual educational material at the same time. Virtual libraries can even be created

where students can study educational material.

Moreover, mobile apps for ﬁnancial education are in abundance at the moment.

Introducing more bite-sized lessons as well as downloadable material that can be

accessed ofﬂine will surely increase the interest of ﬁnance students in M-learning.

However, most of these apps only provide learning material to study. But the number

of apps that provide interactive solutions for students to practice on is near zero.

Therefore, there is scope to make these apps more interactive and enriched with

updated information. Interactive apps will allow students to learn about the stock

market and challenging apps that encourage them to implement managerial decision

skills.

Last but not least, it is true that there is a huge amount of video content on

YouTube and other video portals that provides tutorials on different topics of

ﬁnancial management. Unfortunately, very little video content shows the use of

ﬁnancial management software used by organizations. In the future, more video

creators can contribute to this case.

Discovering the Role of M-Learning Among Finance Students: The Future... 191

8 The Development in Education by Virtue of M-Learning

Mobile learning or m-learning has become ever so popular in recent years. The

spread of mobile devices plays a vital role in this popularity. Since 95% of the human

population lives in an area covered by mobile networks and most adults own more

than one mobile device, it is easy to understand the role and importance of

m-learning in the world today (Crompton & Burke, 2018).

One of the not-so-obvious implementations of m-learning is in education, and this

has become evident in recent times. Before the pandemic, the general population

mostly thought education to be in-person learning. However, the ability of humans

to adapt to any situation has proved this idea to be incorrect and the biggest

contributor which helped prove this concept wrong is m-learning. Not only has

m-learning enabled students to learn from the comfort of their homes or even when

they are on the move, it has also had positive impacts on the students as well. Studies

have found that students perceive collaborative learning positively while learning

through mobile technology (Heﬂin et al., 2017). Other than this, there are many other

implications that mobile technology has on students. The use of mobile technology

has been associated with higher academic performance for students. On top of this,

using mobile technologies for learning can also bring psychological comfort to

students who use their mobile devices all the time. Mobile technology even has

social implications for students, such as integrating education into their lives as a

natural process and not as a training process (Shyshkanova et al., 2017). In general,

mobile technologies increase peer-to-peer engagement and also increase participa-

tion in learning activities (Fabian et al., 2015). M-learning has helped develop not

only the way students perceive education, but also how teachers teach. One research

suggests that mobile learning has a high level of success in project-oriented educa-

tion (Hermann & Gruhn, 2018). M-learning has changed the perspectives of students

and teachers alike, since each new topic presents a new opportunity to learn from a

new angle. One such example of this is the use of mobile technology and augmented

reality to learn Descriptive Geometry (Criollo-C et al., 2018).

9 The Affordability and Availability for Pursuing Studies

as a Finance Student

In the modern age, any student can pursue their studies on the vast global online

education platform that makes education more available and affordable. Finance

students and instructors can access educational materials using digital technology

anytime, anywhere. Students who use information technology no longer have

difﬁculty obtaining learning resources, which are now widely available on the

Internet (Hendra Divayana & Sanjaya, 2017). Finance students and educators also

beneﬁt from online learning platforms because they pursue studies with simple and

quick access to high-quality educational materials; previously, it was only available

in libraries. Since the online application is rapidly developing, students can access

ﬁnance courses, and even ﬁnance students may use affordable mobile devices for

learning. And, according to the research, accessing learning materials from a mobile

device is essential for 64% of learners.

192 A. H. Nadi et al.

Furthermore, 89% of smartphone users download apps, and 50% of students use

apps for educational courses, including ﬁnance courses (Klimova & Polakova,

2020). In foreign language classes, some students may use their mobile devices to

look up terms in translations, either installed or web-based dictionaries. As mobile

devices are effective educational platforms, students can quickly access mobile

devices that provide adequate support for standard Internet technologies. Finance

students can use available and affordable websites to acquire knowledge and calcu-

late necessary transactions through the Internet. Due to the new corona virus disease,

students generally face the problem of textbook affordability, but online platforms

offer a huge opportunity to access e-books. Some Open Education Resource (OER)

sites specialize in a speciﬁc source type, such as textbooks; even the Open Stax and

the Open Textbook Library are two notable textbook available sites (Murphy &

Shelley, 2020). However, ﬁnance students can bring books from the mentioned sites.

The authors suggest that a lack of understanding may hamper the development of

m-learning in Higher Learning Institutions, accessibility to technology tools, and

affordability (Kamaghe et al., 2020). Online education faces various obstacles,

including technological availability and affordability, even when well planned,

including obstacles such as learning differences, as well as the instructors’and

students’technological skills. The growing popularity of mobile applications

requires the banking industry to have a broader view of the market and collaborate

with the FinTech sector (Waliszewski & Warchlewska, 2021).

10 Conclusion

The practical experience is challenging for ﬁnance major students as it includes

ﬁnancial affairs. This chapter ensures the necessity of ﬁnancial classes being

conducted online with effective teaching materials. Numerous personal ﬁnance

software and apps are available online, making ﬁnance students more efﬁcient at

managing money and meeting long-term ﬁnancial goals. As a result, a ﬁnance

student must know about mobile applications available in the market through the

available online courses to prepare for the job market. In addition, ﬁnance students

can improve their ﬁnancial management at home using the right tools. Although it is

affordable to broadcast lectures on a website for many students, online courses

with meaningful interaction among students and instructors are not cost effective

(Baum & Hai, 2020). Digital education provides an opportunity for ﬁnance students

to access ﬁnancial markets using the Internet and gain personal and professional

knowledge. Also, online teaching and learning have been internationalized. For

instance, a well-reputed ﬁnance teacher from the USA may conduct a class online

with the students of another university situated in Asia. As a result, the chapter

ensures that m-learning is signiﬁcant for ﬁnance major students.

Discovering the Role of M-Learning Among Finance Students: The Future... 193

Acknowledgments The Research on Developing Cultural Industry Chain in Guanzhong-Tianshui

Economic Zone (11JK0070), project of the Education Department of Shannxi Provincial Govern-

ment. Studies on Gradient Development of Cultural Industry in Shannxi Province (11E067), project

supported by Social Science Foundation of Shannxi Province, China, The Research on Overall

Design of Cultivating Talents throughout Higher Education (SGH10107), Education Sciences

Planning Project of Shannxi Province, China.

References

Aaron, L. S., & Lipton, T. (2017). Digital distraction: Shedding light on the 21st-century college

classroom. Journal of Educational Technology Systems, 46(3), 1–16. https://doi.org/10.1177/

0047239517736876

Abedin, M. Z., Guotai, C., Moula, F.-E., Azad, A. S. M. S., & Khan, M. S. U. (2019). Topological

applications of multilayer perceptrons and support vector machines in ﬁnancial decision support

systems. International Journal of Finance & Economics, 24(1), 474–507. https://doi.org/10.

1002/ijfe.1675

Abedin, M. Z., Moon, M. H., Hassan, M. K., & Hajek, P. (2021). Deep learning-based exchange

rate prediction during the COVID-19 pandemic. Annals of Operations Research, 2021,

1. https://doi.org/10.1007/s10479-021-04420-6

Abedin, M. Z., Guotai, C., Hajek, P., & Zhang, T. (2022). Combining weighted SMOTE with

ensemble learning for the class-imbalanced prediction of small business credit risk. Complex &

Intelligent Systems. https://doi.org/10.1007/s40747-021-00614-4

Abidin, N. Z., & Tho, S. (2018). The development of an innovative resonance experiment using

smartphones with free mobile software applications for tertiary education. International Journal

of Education and Development using ICT, 14(1), 164–174. Retrieved from https://www.

learntechlib.org/p/183553/

Akour, I., Alshurideh, M., Al Kurdi, B., Al Ali, A., & Salloum, S. (2021). Using machine learning

algorithms to predict people’s intention to use mobile learning platforms during the COVID-19

pandemic: Machine learning approach. JMIR Medical Education, 7(1), e24032. https://doi.org/

10.2196/24032

Aljaaidi, K. S., Bagais, O. A., & Sharma, R. B. (2020). Factors inﬂuencing usage of university

mobile application among university students. The Journal of Asian Finance, Economics, and

Business, 7(10), 1129–1136. https://doi.org/10.13106/jafeb.2020.vol7.no10.1129

Almaiah, M. A., & Alismaiel, O. A. (2018). Examination of factors inﬂuencing the use of mobile

learning system: An empirical study. Education and Information Technologies, 24(1), 885–909.

https://doi.org/10.1007/s10639-018-9810-7

Al-Rahmi, A. M., Al-Rahmi, W. M., Alturki, U., Aldraiweesh, A., Almutairy, S., & Al-Adwan,

A. S. (2021). Exploring the factors affecting mobile learning for sustainability in higher

education. Sustainability, 13(14), 7893. https://doi.org/10.3390/su13147893

Australian Government. (2011). Higher education standards framework. Department of Industry,

Innovation, Science, Research and Tertiary Education. Retrieved from https://www.legislation.

gov.au/Details/F2012l00003/Html/Text#_Toc311791709

194 A. H. Nadi et al.

Baharum, A., Wan, L. Y., Yahya, F., Nazlah, N. H., Nor, N. A., Ismail, I., & Noor, N. A. (2020).

Mobile learning application: Flipped classroom. Indonesian Journal of Electrical Engineering

and Computer Science, 17(2), 1084–1090. https://doi.org/10.11591/ijeecs.v17.i2.pp1084-1090

Baum, T., & Hai, N. T. T. (2020). Hospitality, tourism, human rights and the impact of COVID-19.

International Journal of Contemporary Hospitality Management, 32(7), 2397–2407.

Bazhenov, R. I. (2011). Intelligent information technologies (p. 176). PSU named after Sholem

Aleichem.

Biswas, B., Roy, S. K., & Roy, F. (2020). Students perception of mobile learning during COVID-19

in Bangladesh: University student perspective. Aquademia, 4(2), ep20023. https://doi.org/10.

29333/aquademia/8443

Brain science: The forgetting curve–the dirty secret of corporate training. (2019). Retrieved from

https://learningsolutionsmag.com/articles/1379/brain-science-theforgetting-curvethe-dirty-

secret-of-corporate-training

Camilleri, A. C., & Camilleri, M. A. (2019). Mobile learning via educational apps. In Proceedings

of the 2019 5th International Conference on Education and Training Technologies-ICETT

2019.https://doi.org/10.1145/3337682.3337687

Chansanam, W., Tuamsuk, K., Poonpon, K., & Ngootip, T. (2021). Development of online learning

platform for Thai University students. International Journal of Information and Education

Technology, 11(8), 348–355. https://doi.org/10.18178/ijiet.2021.11.8.1534

Chen, X., Chen, S., Wang, X., & Huang, Y. (2021). “I was afraid, but now I enjoy being a

streamer!”Understanding the challenges and prospects of using live streaming for online

education. Proceedings of the ACM on Human-Computer Interaction, 4(CSCW3), 1–32.

https://doi.org/10.1145/3432936

Chung, E., Subramaniam, G., & Christ Dass, L. (2020). Online learning readiness among university

students in Malaysia amidst COVID-19. Asian Journal of University Education, 16(2), 45.

https://doi.org/10.24191/ajue.v16i2.10294

Connolly, C., Hijón-Neira, R., & Grádaigh, S. Ó. (2021). Mobile learning to support computational

thinking in initial teacher education. International Journal of Mobile and Blended Learning,

13(1), 49–62. https://doi.org/10.4018/ijmbl.2021010104

Criollo-C, S., Lujan-Mora, S., & Jaramillo-Alcazar, A. (2018). Advantages and disadvantages of

M-learning in current education. In 2018 IEEE World Engineering Education Conference

(EDUNINE).https://doi.org/10.1109/edunine.2018.8450979

Criollo-C, S., Abad-Vásquez, D., Martic-Nieto, M., Velásquez-G, F. A., Pérez-Medina, J., &

Luján-Mora, S. (2021). Towards a new learning experience through a mobile application with

augmented reality in engineering education. Applied Sciences, 11(11), 4921. https://doi.org/10.

3390/app11114921

Crompton, H., & Burke, D. (2018). The use of mobile learning in higher education: A systematic

review. Computers & Education, 123,53–64. https://doi.org/10.1016/j.compedu.2018.04.007

Dorouka, P., Papadakis, S., & Kalogiannakis, M. (2020). Tablets and apps for promoting robotics,

mathematics, STEM education and literacy in early childhood education. International Journal

of Mobile Learning and Organisation, 14(2), 255. https://doi.org/10.1504/ijmlo.2020.106179

Fabian, K., Topping, K. J., & Barron, I. G. (2015). Mobile technology and mathematics: Effects on

students’attitudes, engagement, and achievement. Journal of Computers in Education, 3(1),

77–104. https://doi.org/10.1007/s40692-015-0048-8

Faizi, R., El Aﬁa, A., & Chiheb, R. (2013). Exploring the potential beneﬁts of using social media in

education. International Journal of Engineering Pedagogy (iJEP), 3(4), 50. https://doi.org/10.

3991/ijep.v3i4.2836

Fombona, J., Pascual, M. A., & Pérez Ferra, M. (2020). Analysis of the educational impact of

M-learning and related scientiﬁc research. Journal of New Approaches in Educational

Research, 9(2), 167. https://doi.org/10.7821/naer.2020.7.470

Greenhow, C., & Lewin, C. (2019). Social media and education: Reconceptualizing the boundaries

of formal and informal learning. Social Media and Education, 2019,6–30. https://doi.org/10.

4324/9781315121697-2

Discovering the Role of M-Learning Among Finance Students: The Future... 195

Guy, R. (2009). The evolution of mobile teaching and learning. Informing Science.

Hamilton, E. (2019). Why mobile apps are important for your business? Retrieved from https://

www.techtimes.com/brandspin/242588/20190504/why-mobile-apps-areimportant-for-your-

business.htm

Hasan, M., Noor, T., Gao, J., Usman, M., & Abedin, M. (2022). The impact of ﬁnancial knowledge

on Fintech access: Client perspective evidence from Bangladesh. Journal of the Knowledge

Economy.

Heﬂin, H., Shewmaker, J., & Nguyen, J. (2017). Impact of mobile technology on student attitudes,

engagement, and learning. Computers & education, 107,91–99.

Hendra Divayana, D. G., & Sanjaya, D. B. (2017). Mobile phone-based cipp evaluation model in

evaluating the use of blended learning at school in Bali. International Journal of Interactive

Mobile Technologies, 11(4).

Hermann, J., & Gruhn, V. (2018). M-learning to support project-oriented higher education in

software engineering. In New trends in intelligent software methodologies, tools and techniques

(Vol. 303, pp. 705–712). https://doi.org/10.3233/978-1-61499-900-3-705

Herrador-Alcaide, T. C., Hernández-Solís, M., & Hontoria, J. F. (2020). Online learning tools in the

era of M-learning: Utility and attitudes in accounting college students. Sustainability, 12(12),

5171. https://doi.org/10.3390/su12125171

Hossain, S. F. A., Nurunnabi, M., & Hussain, K. (2021). Continuous mobile devices usage

tendency in the TPACK-based classroom and academic performance of university students.

Technology, Pedagogy and Education, 30(4), 589–607. https://doi.org/10.1080/1475939X.

2021.1933160

Hossain, S. F. A., Xi, Z., Nurunnabi, M., & Anwar, B. (2022). Sustainable academic performance in

higher education: A mixed method approach. Interactive Learning Environments, 30(4),

707–720. https://doi.org/10.1080/10494820.2019.1680392

Jaber, K., Abduljawad, M., Ahmad, A., Abdallah, M., Salah, M., & Alhindawi, N. (2021).

E-learning mobile application evaluation: Al-Zaytoonah University as a case study. Interna-

tional Journal of Advances in Soft Computing and its Applications, 13(3), 89–99. https://doi.

org/10.15849/ijasca.211128.07

Kacetl, J., & Klímová, B. (2019). Use of smartphone applications in English language learning—A

challenge for foreign language education. Education Sciences, 9(3), 179. https://doi.org/10.

3390/educsci9030179

Kamaghe, J. S., Luhanga, E. T., & Michael, K. (2020). The challenges of adopting M-learning

assistive technologies for visually impaired learners in higher learning institution in Tanzania.

Khan, M. N. R., Yesmin, S., Aktar, M., Chowdhury, K. B. Q., Labeeb, K., & Abedin, M. Z. (2021).

Techniques for multi-omics data incorporating machine learning and system genomics. In 2021

6th International Conference on Communication and Electronics Systems (ICCES)

(pp. 1524–1528). IEEE.

Khan, M., Nasﬁkur, R., Shakir, A. K., Nadi, S. S., & Abedin, M. Z. (2022). An android application

for university-based academic solution for crisis situation. In Sentimental analysis and deep

learning (pp. 653–663). Springer.

Klimova, B. (2019). Impact of mobile learning on students’achievement results. Education

Sciences, 9(2), 90. https://doi.org/10.3390/educsci9020090

Klimova, B., & Polakova, P. (2020). Students’perceptions of an EFL vocabulary learning mobile

application. Education Sciences, 10(2), 37.

Kumar Basak, S., Wotto, M., & Bélanger, P. (2018). E-learning, M-learning and D-learning:

Conceptual deﬁnition and comparative analysis. E-Learning and Digital Media, 15(4),

191–216. https://doi.org/10.1177/2042753018785180

Laurens Arredondo, L. A., & Valdés Riquelme, H. (2021). M-learning adapted to the ARCS model

of motivation and applied to a kinematics course. Computer Applications in Engineering

Education. https://doi.org/10.1002/cae.22443

196 A. H. Nadi et al.

Llerena-Izquierdo, J., Barcia-Ayala, O., & Ayala-Carabajo, R. (2020). Faculty training through

crowdlearning for emerging online education. In 2020 IEEE ANDESCON.https://doi.org/10.

1109/andescon50619.2020.9272103

Mishra, L., Gupta, T., & Shree, A. (2020). Online teaching-learning in higher education during

lockdown period of COVID-19 pandemic. International Journal of Educational Research

Open, 1, 100012. https://doi.org/10.1016/j.ijedro.2020.100012

Murphy, J. A., & Shelley, A. (2020). Textbook affordability in the time of COVID-19. Serials

Review, 46(3), 232–237.

Naciri, A., Baba, M. A., Achbani, A., & Kharbach, A. (2020). Mobile learning in higher education:

Unavoidable alternative during COVID-19. Aquademia, 4(1), ep20016. https://doi.org/10.

29333/aquademia/8227

Nguyen, D. C., Ding, M., Pathirana, P. N., Seneviratne, A., Li, J., Niyato, D., Dobre, O., & Poor,

H. V. (2022). 6G internet of things: A comprehensive survey. IEEE Internet of Things Journal,

9(1), 359–383. https://doi.org/10.1109/JIOT.2021.3103320

Oremus, B. (2021, December 30). In 2021, tech talked up ‘the metaverse.’One problem: It doesn’t

exist. Retrieved from https://www.washingtonpost.com/technology/2021/12/30/metaverse-

deﬁnition-facebookhorizon-worlds/

Papadakis, S., & Kalogiannakis, M. (2017). Mobile educational applications for children: What

educators and parents need to know. International Journal of Mobile Learning and Organisa-

tion, 11(3), 256. https://doi.org/10.1504/ijmlo.2017.085338

Pedro, L. F., Barbosa, C. M., & Santos, C. M. (2018). A critical review of mobile learning

integration in formal educational context. International Journal of Educational Technology in

Higher Education, 15(1). https://doi.org/10.1186/s41239-018-0091-4

Pratama, A. R. (2020). Fun ﬁrst, useful later: Mobile learning acceptance among secondary school

students in Indonesia. Education and Information Technologies, 26(2), 1737–1753. https://doi.

org/10.1007/s10639-020-10334-w

Shyshkanova, G., Zaytseva, T., & Frydman, O. (2017). Mobile technologies make education a part

of everyday life. Information and Learning Science, 118(11/12), 570–582. https://doi.org/10.

1108/ils-03-2017-0019

Singh, R., Timbadia, D., Kapoor, V., Reddy, R., Churi, P., & Pimple, O. (2021). Question paper

generation through progressive model and difﬁculty calculation on the Promexa mobile appli-

cation. Education and Information Technologies, 26(4), 4151–4179. https://doi.org/10.1007/

s10639-021-10461-y

Slavova, Y., & Mu, M. (2018). A comparative study of the learning outcomes and experience of VR

in education. In 2018 IEEE Conference on Virtual Reality and 3D User Interfaces (VR).https://

doi.org/10.1109/vr.2018.8446486

Soﬁana, N., & Mubarok, H. (2020). The impact of English game-based mobile application on

students’reading achievement and learning motivation. International Journal of Instruction,

13(3), 247–258. https://doi.org/10.29333/iji.2020.13317a

Talan, T. (2020). The effect of mobile learning on learning performance: A meta-analysis study.

Educational Sciences: Theory & Practice, 20(1), 79–103. https://doi.org/10.12738/jestp.2020.

1.006

Trinh, L. T., Thao, T. T., Hang, T. T., Thanh, N. C., & Trung, T. (2021). Analysis of students’

ability to accept M-learning technology: An exploratory study from high schools in Vietnam.

International Journal of Interactive Mobile Technologies (iJIM), 15(12), 86. https://doi.org/10.

3991/ijim.v15i12.22143

Wahyudi, W., Tukan, B. A. P., & Pinem, D. (2020). Analysis of the effect of ﬁnancial literation,

ﬁnancial technology, income, and locus of control on lecturer ﬁnancial behavior. AFEBI

Management and Business Review, 5(1), 37–46.

Waliszewski, K., & Warchlewska, A. (2021). How we can beneﬁt from personal ﬁnance manage-

ment applications during the Covid-19 pandemic? The polish case. Entrepreneurship and

Sustainability Issues, 8(3), 681.

Zhylenko, T. I., Martynova, N. S., Shuda, I. A., Chykalov, Y. A., & Kuzmuk, D. A. (2020). Auto

checker of higher mathematics –An element of mobile cloud education. Retrieved from http://

elibrary.kdpu.edu.ua/bitstream/123456789/3895/1/paper39.pdf

197

Exploring the Role of Mobile Technologies

in Higher Education: The Impact of Online

Teaching on Traditional Learning

Syed Far Abid Hossain, Armana Hakim Nadi, Rahma Akhter,

Md. Ahmedul Islam Sohan, Faiza Tanaz Ahsan, Mahbuba Rahman Shoﬁn,

Saadmann Shabab, Tanusree Karmoker, and Krishna Paul

Abstract The chapter aims to explore the role of mobile technologies in higher

education especially the impact of online teaching on traditional learning. The

transformation of the educational setting from online to ofﬂine draws limited

attention from researchers in the post-pandemic era. The key reason for conducting

this chapter is to explore the hidden issues of student coping strategies in the ofﬂine

learning environment. In addition, the chapter explores the opportunities and limi-

tations of technology usage in higher education. The study utilized a qualitative

research approach to conduct the chapter with an extensive literature review. The

result shows that with the advanced usage of mobile technology, the academic

resources are freely available and accessible to all the learners that can ensure

effective teaching and learning, however, the study is conducted among a limited

number of respondents in a single country. This may affect the generalization of the

study.

Keywords Mobile technologies · Higher education · Online teaching · Traditional

learning

S. F. A. Hossain (✉) · R. Akhter

BRAC Business School, BRAC University, Dhaka, Bangladesh

e-mail: rahma.akhter@bracu.ac.bd

A. H. Nadi

Bangladesh University of Professionals, Dhaka, Bangladesh

M. A. I. Sohan · M. R. Shoﬁn · T. Karmoker

IUBAT University, Dhaka, Bangladesh

e-mail: asohan@iubat.edu

F. T. Ahsan · S. Shabab · K. Paul

North South University, Dhaka, Bangladesh

e-mail: faiza.tanaz@northsouth.edu

M. Z. Abedin, P. Hajek (eds.), Novel Financial Applications of Machine Learning

and Deep Learning, International Series in Operations Research & Management

Science 336, https://doi.org/10.1007/978-3-031-18552-6_12

198 S. F. A. Hossain et al.

1 Introduction

Mobile technologies have enormous potential to transform education if they are

developed and applied in a way that is appropriate for the social and cultural

environment in which students learn with up-to-date technology. In the era of

learning with industry 4.0 (Yang et al., 2022), mobile technology’s use, implemen-

tation, and design in the higher educational setting provide technological and

sociocultural obstacles. Various studies have shown that in nations such as the

USA, the United Kingdom, Germany, France, and Japan, there are more 5G

internet-capable mobile phones (Nguyen et al., 2022)with comparable functionality

than genuine desktop computers (Rmenardi, 2012) that enhanced the learning

trajectory such as deep learning (Abedin et al., 2021). Mobile technologies are

described as all technological devices that are portable and lightweight (Lai et al.,

2022) that can connect to the Internet via wireless connections or data cables, such as

smartphones, iPads, and PDAs. M-learning is also deﬁned as a dynamic learning

environment enabled by the use of mobile technologies, especially in the ﬁelds of

education (Keengwe & Bhargava, 2013). Given the digital environment of the

twenty-ﬁrst century, the application of mobile technology to education is essential

to investigate how these applications change the social structure of learning envi-

ronments in different learning environments, as well as how mobile technologies

shape learning environments. By addressing different learning styles of learners and

providing educational materials to everyone, anywhere, anytime, and in various

versatile formats such as podcasts, audio recordings, or videos, mobile technologies

can be of great help to education, strategy, organization, and content. Students and

trainees working in distant ﬁeld regions can communicate with their lecturers and

obtain information via mobile devices from anywhere and at any time. Patients can

beneﬁt from mobile technology when used for notiﬁcations, reminders, language

acquisition, motivation, and guiding. As a result, mobile technologies can provide a

portable, lightweight learning platform that can result in private and spontaneous

learning (Traxler, 2005). Mobile phones’IM (Instant Messaging) capabilities can aid

in the creation of learning environments that improve knowledge transformation

(Kekwaletswe, 2007). We have reached the mobile era, in which people carry their

mobile gadgets with them at all times. Mobile technologies offer the potential to

promote informal education from anywhere, at any time, and in any context. The

major focus should be placed on recognizing that new learning applications arise

through interaction and communication among the main participants in the devel-

opment cycle and that mobile technologies are facilitating technology (Sharples,

2007). The development of modern society requires well-educated people. Mobile

technologies have the potential to turn education into a seamless aspect of everyday

life, to the point that people no longer identify it as training. The learning process

will become natural and easy and the quality of learning will improve (Shyshkanova

et al., 2017). The advancement of wireless technology in education, as well as the

development of mobile apps, is astounding. Mobile technology in education has

become one of the most signiﬁcant areas of research and application in recent years.

For many educational institutions, mobile learning is becoming a crucial concern.

Because new types of devices and apps are transforming education, it is critical to

ensure that mobile learning is properly used and implemented (Sattarov & Khaitova,

2019).

Exploring the Role of Mobile Technologies in Higher Education: The... 199

Recent literature discovered diversiﬁed phenomena such as mobile applications

to utilize ﬁnancial decision support system (Abedin et al., 2019), continuous trend of

smartphone usage in collaboration with TPACK-based lesson plan development

(Hossain et al., 2021), evaluation of the FinTech opportunity for the organization

with updated technological advancement (Hasan et al., 2022), sustainable academic

performance in higher education with cutting-edge technology of smartphone in

higher education (Hossain et al., 2022), complex and intelligence system develop-

ment (Abedin et al., 2022), and many more; however, the impact of online teaching

and learning (Hossain et al., 2019) on traditional teaching and learning style is still

under shadow.

2 Literature Review on Mobile Technologies in Teaching

The debate about the use of technology in education dates back at least 2500 years.

To better comprehend the role and impact of technology on education, we need to go

back in time, because there are always lessons to be learned from history. One of the

most comprehensive historical histories is Paul Saettler’s“The Evolution of Amer-

ican Educational Technology”in 1990; however, it only covers up to 1989. Since

then, a lot has transpired. Teemu Leinonen has a wonderful research article on recent

history as well (Leinonen et al., 2010) (Fig. 1).

During the 1990s, the expense of making and dispersing video dropped signiﬁ-

cantly because of computerized pressure and rapid Internet access. This decrease in

the expenses of recording and appropriating video likewise prompted the improve-

ment of talk catch frameworks. The innovation allows understudies to view or audit

addresses whenever and place with an Internet association. YouTube began in 2005

and is progressively being utilized for short instructive clasps that can be

downloaded and coordinated into online courses. It is also seen that The Khan

Academy began using the YouTube platform in 2006 for recorded voice-over

addresses involving an advanced chalkboard for conditions and delineations.

Apple Inc. made iTunes U in 2007 to turn into a gateway or a webpage where

recordings and other computerized materials on college instruction could be gath-

ered and downloaded for nothing by end clients.

Technology puts students on the way to customizing learning by giving them the

power to control their studies, make education relevant to their digital lives, and

prepare them for their future. Students are driven to become reﬂective practitioners,

collaborators, creators, and critical thinkers as a result of access to technologies and

resources outside the classroom. When technology is well integrated into the class-

room, students have a lifetime of learning love (Arnold & Sangrà, 2018). Instructors

are always working to customize learning for their students. Technology can help

them reach new heights by accessing student data in real time, transverse informa-

tion, materials, applications, and more. Software can improve teachers’efforts in

creating hybrid learning environments and using digital tools for summative and

formative evaluations, introducing new paradigms of teaching and learning into

classrooms.

200 S. F. A. Hossain et al.

Printer

Internet

Laptop

Tablet PC

Mobile

Technologies

Teaching

Printing with

shared

connection

Internet

connectivity

with Laptop

Integration of

Laptop and

smartphone

Integration of PC

and smartphone

and Tab

Laptop

and Tab

Laptop

and PC

MOOCs

m-learning

e-learning

distance learning

Flipped learning

VR learning

AR learning

mechine learning

Artificial

intelligence

Collaborative

learning

Fig. 1 Technologies in teaching (Source: Author’s own creation)

A study by Intel Corporation suggests that digital learning, as well as having the

correct devices in students’hands, helps them prepare for the professional life and

provide them with technical skills needed by the future workforce. Relevant STEAM

learning experiences can provoke creativity, help students apply meaning to learn-

ing, and plan future career opportunities and undeveloped careers. Physical compu-

tation, coding, programming, and computer thinking skills are common in this

profession. Students may learn these techniques while also improving their critical

thinking and problem-solving skills for the twenty-ﬁrst century through the creation.

Design and proper technology can make learning with manufacturers and the

environment very stimulating. School and universities face difﬁculties in deciding

which devices and technologies will help them realize their ambitions of changing

learning. Working with various stakeholders to evaluate how teachers and students

use devices for daily learning, devices should be used to select devices. Stakeholders

must consider acceptable content requirements, grade-level curricula, and how

devices will be used. It is not a simple chore, but factors like assessment needs,

security features, compatible digital curriculum and material, management choices,

device performance, and total cost of ownership all play a role in selecting the

correct device. The basis of a 360-degree learning experience is a safe and strong IT

infrastructure that supports digital material, protects important student data,

increases operational efﬁciency, and ensures safety of the students.

Exploring the Role of Mobile Technologies in Higher Education: The... 201

Besides the development and use of virtual classrooms and online-based educa-

tion platforms, online media are actually a subclass of PC innovation; however, their

improvement merits its very own segment throughout the entire existence of instruc-

tive innovation. Web-based media cover a wide scope of various advancements,

including websites, wikis, YouTube recordings, cell phones like telephones and

tablets, Skype, Facebook, and Twitter. Kaplan & Haenlein, (2010) characterize

web-based media collectively “of Internet based applications that permit the crea-

tion and trade of client produced content, in light of cooperations among individuals

in which they make, offer or trade data and thoughts in virtual networks and

organizations.”

The gap in the past literature reviews that we are going to address in this paper is

the substitutability of online classes with a physical classroom-based study session

due to the prevalence of COVID-19 since early 2020. The tsunami of web-based

learning has occurred. Many schools offer on the Web (virtual) learning for under-

studies as a method of continuing education during the remainder of the school year.

Educationalists and directors who hesitantly teach on the Internet have only a few

choices to accept the decade-old innovation. Some instructors may encounter fears

and fear when moving their home room to the Internet, but most of them do so

quickly and within a short period of time; over the long haul, everybody appears to

adjust well. The advanced separation is more obvious than ever in recent memory

(Guernsey et al., 2020). Children who can bring computers are ready. Educators and

showing strategies are a piece behind; notwithstanding, there is conﬁdence in the

creation of a new school model. Change can be valuable.

3 The Inﬂuence of Mobile Technologies in Teaching

In today’s world, most of the population keeps smartphones in their possession at a

very early age (Han, 2022). It goes without saying that mobile technologies are used

for much more than just communication. In fact, mobile technology is one of the

most recent tools to support real-world learning (Hashim, 2018).

Like any other technology available in the world in the contemporary era, mobile

technologies are no different in terms of inﬂuencing users and stakeholders both

positively and negatively, especially with hedonic usage (Vujić& Szabo, 2022).

Research shows that mobile technologies are associated with a positive perception of

students in collaborative learning, but that students are more dissatisﬁed in class

(Heﬂin et al., 2017). Positive inﬂuences of modern technology on education include:

globalization and improvements in education and learning without geographical

restrictions. In contrast, negative inﬂuences include: increasing incidents of

cheating, declining writing skills, and lack of focus (Raja & Nagasubramani, 2018).

202 S. F. A. Hossain et al.

Currently, especially in this post-COVID era, the usage of mobile technology has

become part and parcel of education. Mobile technologies have facilitated improved

means of education through increased portability and easy access to the Internet.

Now teachers and students can search for a topic and learn on-the-go. Mobile

devices enable students to easily access education content from any place and at

any time (Criollo-C et al., 2018). Students are able to tutor themselves through video

tutorials or downloadable bite-sized lessons from the Internet. Through the means of

online education facilitated by smartphones apps, students around the world are now

able to familiarize themselves with international contexts. Furthermore, students

hailing from different areas of the world are now able to attend online classrooms at

the same time through virtual meeting apps such as Zoom, Google Meet, etc., which

greatly reduces geographical barriers. Students of the modern era can easily com-

municate and enhance their network on a global scale through mobile technologies.

However, there are certain negative impacts to this facility. It is true that mobile

technologies have facilitated on-the-go learning, but this also means that anyone

anywhere can search on topics they want to learn or, if they intend to, copy in their

exams. Some students may tend to use unfair means in their examination through

mobile technology. Moreover, depending on the texts and material composed by

others is also greatly reducing the creativity of students. Instead of coming up with

their own ideas, students are becoming dependent on information which is already

available through online media. This, in turn, also results in the lack of focus of

students and the creation of a mindset among them to depend on online educational

materials without giving much concentration in classroom sessions.

4 Mobile Technologies Available via an Online Platform

Even in the recent past, smartphones were considered hazardous to the educational

well-being of students, and parental control seems very strict according to the

existing literature (Hadad et al., 2020). However, this scenario has changed to a

great extent at present. Utilizing new innovations in technologies, smartphone apps

have contributed signiﬁcantly in the ﬁeld of education. At the moment, there are

hundreds, if not thousands, of mobile apps providing educational support to stu-

dents, scholars, and teachers from around the world. Moreover, most of these

educational apps are free. Among the most notable free educational apps, the ten

most prominent ones are mentioned in the table below (Mindster, 2020) (Table 1).

Apart from the mobile apps mentioned in the table above, other mobile technol-

ogies are also available that contribute a lot to the ﬁeld of education. Figure 2

represents various income groups with life expectancy. The overall income or

GDP is an indicator of the use of individually owned technological devices.

According to life expectancy data, the use of technology in the classroom may

vary signiﬁcantly. For example: the use of cloud-based Learning Management

Systems such as Moodle, Blackboard, etc., in combination with web conferencing

platforms such as Zoom, Google Meet, etc. has revolutionized the education sector.

Description Source

Both students and teachers are enjoying the beneﬁts of online education through

these services.

Exploring the Role of Mobile Technologies in Higher Education: The... 203

Table 1 Various mobile apps for educational purposes available on the online platform

Mobile

App

Google

classroom

A virtual classroom that facilitates submitting and grading

assignments, sending announcements, starting discussions, cre-

ating classes, sharing resources, asking for remarks and answers,

and so on.

Tarteer et al.

(2021)

edX Educational material from top universities such as Harvard, MIT,

Columbia, etc. including compilation of more than 2000 courses

like engineering, computer science, linguistics, business studies,

and many more.

Shi and Lin

(2021)

Khan

academy

Platform providing lessons in the form of video tutorials. The

video shows the drawing recorded on the virtual black board that

the narrator shows. Khan academy also offers online courses to

prepare standardized exams such as SAT, MCAT, and LSAT.

Massey et al.

(2022)

Duolingo Language learning app that facilitates learning of 30+ languages

in an interactive way through mini games. The app also tracks

the performance of the learner and provides insight.

Ahmed et al.

(2022)

Remind Community that helps students learn in groups and stay

connected. Remind is also used to message the entire class,

submit assignments, share photos and handouts, and clear doubts

with friends, individually and in collaboration.

Jones et al.

(2022)

Photomath Solves mathematical problems by providing step-by-step expla-

nations and instructions to the learner by utilizing submitted

photos; either handwritten or printed.

Long and

Bouck (2022)

SoloLearn A platform providing tutorials for learning coding languages

such as C++, Java, python, swift, JavaScript, CSS, PHP, HTML,

and so on.

Elsisi et al.

(2022)

Quizlet Simple tools that help students practice and master any topics

they prefer. Quizlet allows learners to design their own sets or

gather sets from other contributors and study them.

Senior (2022)

Kahoot Provides ready-made quizzes on any topic of interest. Learners

can take individual quizzes or participate in live quizzes with

other learners.

Vijayakumar

(2022)

uDemy Holds more than 130,000 video tutorials for courses ranging

from business and technology to personal development. If the

learner feels stuck in a particular lesson, he or she can ask

questions to other students and teachers and solve their doubts.

Moslehi et al.

(2022)

5 Popular Applications in Higher Education

Higher education institutions have begun to use mobile technologies to improve

education quality (Han & Shin, 2016). Although institutions do not only improve the

quality of higher education, they also assist students in learning. Numerous tools and

applications are available to imply the overall higher education system. The most

common applications are such as Virtual reality applications (Radianti et al., 2020),

Game-based learning applications (Subhash & Cudney, 2018), Blockchain-based

applications (Awaji et al., 2020), and so on.

204 S. F. A. Hossain et al.

Fig. 2 Various income groups with life expectancy

However, the twenty-ﬁrst century learning aid is virtual reality (Rogers, 2019).

According to one study, after participating in VR activities, students can retain more

information and better apply the information they have learned (Krokos et al., 2019).

The authors suggest that integrated gameplay features are more efﬁcient in increas-

ing learner engagement (Scholz et al., 2021). The game-based application can help

students assess learners and personalized collaboration in group playing, enhancing

the level of knowledge (Troussas et al., 2020). Students and educational institutions

can also use a blockchain-based application to create more personalized digital

agreements that include assignment criteria, time frames, and grading procedures

(Chen et al., 2018). Despite that, EduCTX is a blockchain-based decentralized

higher education credit network (Turkanovic et al., 2018). These systems are

ﬂexible, secure, and reliable due to their global storage capacity and resource sharing

(Coulouris et al., 2012). According to research, most higher education institutes offer

online courses to their students through websites, learning platforms, video confer-

encing (Google Meet, Google Duo, Google Hangout, and Zoom), YouTube, Social

media (Facebook and Twitter), and several other free websites for blended learning

tools (Shahzad et al., 2020). Furthermore, emerging technology improved several

aspects of student participation in the three dimensions of engagement, with web

conferencing software, digital games, and Facebook seem to be the most signiﬁcant

(Schindler et al., 2017). The features of mobile technology, such as portability and

mobility, improve their potential application in education through the use of digita-

lized library and information access; many higher education institutions worldwide

are exploring its possible use in higher education (Yip et al., 2020). It can remove

space and time limitations to education, thereby greatly expanding participation in

higher education. Therefore, online application has become a necessary and ongoing

investment in the modern era, and most educational institutions must gather addi-

tional funds to fulﬁll their expanding needs for technology resources (Baldwin,

2021). There is little doubt that technology will be one of the driving forces in the

development of higher education. In existing models, technology should be consid-

ered a component that impacts student involvement. Soon, the development of

educational apps will be a once-in-a-lifetime opportunity for developers.

Exploring the Role of Mobile Technologies in Higher Education: The... 205

6 Higher Education from Online to Ofﬂine Setting

After the global pandemic has stabilized, the education system will move from

online to ofﬂine platforms. In an ofﬂine setting, professors have spent a signiﬁcant

portion of their class time with students distributing material through lectures and

follow-up discussions. As a large group of participants, universities have been forced

to take similar preventive measures to minimize the impact of COVID-19 on higher

education and beyond (Liguori & Winkler, 2020). The current increase in education

is the technology to combine teaching information online and ofﬂine. However,

when higher education institutions adopted online learning settings, they faced

challenges, but when considering online learning to ofﬂine, it seems back to the

traditional learning process. The authors investigated the high combination of digital

technologies and academic teaching to grow students’autonomous learning ability

and increase classroom vitality, which is valuable and relevant for improving

classroom teaching (Chen et al., 2020). Higher education in online learning has

formal and informal approaches; on the other hand, ofﬂine learning is formal to

increase its role as a learning platform to provide students with various learning

opportunities. Ofﬂine teaching increases active interaction between teachers and

students, even though online learning allows students to study solely via the Internet

on their own. In the Covid-19 pandemic and post-epidemic era, the teaching model

combines the beneﬁts of online and ofﬂine learning in higher education (Gao & Lu,

2021). Currently, online practices are connected to ofﬂine instruction to better

understand learning materials. According to a study, online teaching is ineffective

than face-to-face learning (Liu & Han, 2020). In addition, transformation methods

from online to ofﬂine enhance student communication with the instructor, active

learning, and reduce procrastination.

206 S. F. A. Hossain et al.

7 The Impacts of Mobile Technologies on University

Students

Mobile technologies have come a long way since their inception. In the past, people

used mobile phones that would be weighted like a dumbbell. However, thanks to the

advancement in mobile technology, people now use phones that ﬁt in the palm of

their hands and have bodies that are mostly screens. Mobile technology affects

people of all ages but in the recent times the group of people that it has had the

most effect on are students. Studies have found that students perceive collaborative

learning positively while learning through mobile technology (Heﬂin et al., 2017).

Other than this, there are many other implications that mobile technology has on

students. The use of mobile technology has been associated with higher academic

performance of students. On top of this, using mobile technologies for learning can

also bring psychological comfort to students who use their mobile devices all the

time. Mobile technology even has social implications on students, such as integrat-

ing education into their lives as a natural process and not as a training one

(Shyshkanova et al., 2017). In general, mobile technologies increase peer-to-peer

engagement and also increase participation in learning activities (Fabian et al.,

2015). However, not all the impacts that mobile technologies have are positive.

There are many negative impacts that mobile technology can have on students.

Mobile technology, while increasing student engagement, can also be the cause of

distraction. Since students access their social lives mainly on their mobile devices, it

is quite easy for them to become distracted with social media while trying to engage

in any learning activities. Another negative impact mobile technology can have on

students is that it can increase the number of students using unfair means on tests and

assessments since they have full access to the Internet and no one to supervise them.

Another study found that the more mobile technology is used for social interaction,

the more it negatively affects the quantity and quality of face-to-face interactions

(Elsobeihi & Abu Naser, 2022). In other words, due to too much reliance on mobile

technology for social interaction, users become more and more disengaged to

in-person social interaction.

8 The Impacts of Variation in Assessment in Higher

Education

Covid-19 has caused changes in almost all aspects of everyone’s lives. From how

people work to how people commute. This pandemic has once again proven just how

adaptable humans are. However, one of the largest industries that Covid-19 has

impacted is the education industry. The education sector has changed enormously

toward E-learning and M-learning. Although most of the impact has been positive,

there are still some negative impacts that M-learning has had on education. One such

area where M-learning has had both a positive and a negative impact is in the

assessment process. One study states that lack of preparation and the inherent

downsides of remote assessment have proven to be an extraordinary challenge for

higher education assessment. Some of these challenges include dishonesty, the lack

of proper infrastructure, submission deadline commitment, and so on (Guangul et al.,

2020). Since teachers and faculties do not have any way to observe their students live

in a controlled environment, there are high chances that students use unfair and

dishonest means to complete their assessments. Although assessment technology has

come very far in just a couple of years, it is yet to be considered as the solution which

can completely eliminate dishonest means in assessments. Again, it is not the case

that the students are always at fault. Sometimes students are the one who fall victim

to the lack of proper infrastructure, leading to late submission or even missed

assessments.

Exploring the Role of Mobile Technologies in Higher Education: The... 207

However, not all the impacts of online assessments have been negative. Mobile

learning has enabled students and teachers both to access the assessments at their

convenience time and place. Teachers also face challenges when it comes to online

assessments. Since in M-learning there is no face-to-face interaction, teachers ﬁnd

difﬁculty in conveying their intentions (Kearns, 2012). Another substantial problem

that is common with online assessment is the risk that students will get the assess-

ment responses in advance. This can be caused by various factors like faulty

infrastructure, hacking, and even dishonest assistants. Of course, since everyone

takes online assessments in their convenient time, thus students can easily share the

answers with their peers taking the assessment in a different time than theirs (Rowe,

2004). All in all, online assessment is the part of M- or E-learning that needs to be

developed the most, and although there are certain advantages to online assessments,

the disadvantages of online assessments overshadow them.

9 Traditional, Online, or Blended Learning?

Traditional learning is a face-to-face interaction process; through this way, there is

no need to worry about security and conﬁdentiality issues in traditional education, as

in the case with online education; as well as in this process, a student of higher

education can gain connection, inspiration, availability, structure, and so on

(Razeeth et al., 2019). Also, connectivity leads to direct communication between

students and professors so that higher education as a consequence students are

enabled to expand their collaborative activities and eliminate direct doubts about

speciﬁc issues promptly, which is different from online learning. Furthermore, most

of the time in traditional learning professors present and discuss topics; on the other

hand, students pay close attention and try to understand the topics (Azzalis et al.,

2009). Various scholarly articles indicate that this way of learning improves stu-

dents’capacity to recall and grasp new content (Hyun et al., 2017). But in the age of

technological advancement, students want to be able to read material from any-

where, and it has become possible through online learning in higher education. Due

to this need, online education has become an effective and desirable choice. Online

learning is becoming highly popular among students in higher education, as well as

they believe that the traditional learning format is rigid, authoritarian, and

unsustainable and higher education may now provide efﬁcient classroom instruction

through the Web in this advancing age (Paul & Jefferson, 2019).

208 S. F. A. Hossain et al.

For higher education, students want to have a better education without having to

abandon employment, home life, or transportation costs. In addition, online learning

students have the opportunity to contact professors, engage friends and classmates,

study documents, and ﬁnish all the class tasks through any Internet accessible point,

rather than needing to be in a given place at a particular time frame (Richardson &

Swan, 2003). As online learning is growing in popularity, various higher education

institutions are fond of determining the best way to distribute course content among

online students (Dumford & Miller, 2018). As a result, higher education institutions

have begun to embrace mobile technologies to meet student requisites (Han & Shin,

2016). The ﬁrst and foremost reason for learning online these days is the assault of

the Covid-19 virus, which has led to large-scale migration from traditional face-to-

face learning to online learning. Millions of teaching members across the world

began lecturing in front of electronic screens shortly after the start of 2020, while

their pupils were required to remain at home and attend courses over the Internet

(Bao, 2020). Another thing is that online learning is more ﬂexible than traditional

learning. In response to fears about the rapid spread of the coronavirus around the

world, a large number of educational institutions around the world have temporarily

stopped face-to-face classes to prevent it from spreading, leading universities around

the world to shift more toward online learning, and other research authors have also

suggested online and distance education as a necessity during social distance with

lockdown due to the COVID-19 pandemic (Ali, 2020). The coronavirus has also

shown new threats to the entire education system, demonstrating that society needs a

reliable and versatile education system to confront an uncertain future. Another

learning term is blended learning (BL), which combines traditional face-to-face

learning alongside online learning, is a technological advancement that is drastically

revolutionizing teaching and learning in higher education, and is becoming more

popular in higher education. And blended learning is often used in a combination of

phrases that include merged ﬂexible, mixed mode, or hybrid learning (Anthony

et al., 2019). Previous research tested the efﬁcacy of blended learning by comparing

traditional and online teaching, as there has been tremendous progress in blended

learning that has emphasized improving learning and teaching outcomes (Van Laer

& Elen, 2020). Online activities such as wordbooks, study guides, online writing

tools, discussion forums, web addresses, video tutorials, relevant materials, models,

exercises, quizzes, and so on are all part of the layout and execution of blended

learning online educational materials (Anthony et al., 2019). Inversely, traditional

face-to-face education includes lectures, laboratory activities, face-to-face practice

and skills assessment, individual/group presentations, and professor-led discussions

to assess students’academic performance (Sun & Qiu, 2017).

According to the results of a previous research paper, blended learning methods

improve the acquisition of knowledge, learning engagement, and wisdom because it

has a remarkable impact on the consciousness and learning backgrounds of students

and emphasizes learning from blended learning (Edward et al., 2018), thus guiding

students in becoming more engaged in the learning process and allowing them to be

more enthusiastic, which enhances their patience and dedication (Ghazal et al.,

2018). Blended Learning uses a blend of online and traditional face-to-face (F2F)

learning to assist professors in achieving educational goals in higher education

students, to build efﬁcient and productive logical knowledge, help improve educa-

tional aspects, and establish social discipline (Subramaniam & Muniandy, 2019).

Keeping in mind student and lecturer perspectives, blended learning works to

establish a peaceful, coherent equilibrium, prosperous, and healthy combination

among online information availability and traditional learning in higher education

(Bervell & Umar, 2018). A previous research also mentioned that blended learning

comprises a combination of several activities, which is achieved by integrating 70%

online learning and 30% face-to-face engagement (Anthony et al., 2019). Students’

enthusiasm in their learning path grows as a result of blended learning (Chang-Tik,

2018), allows students to learn at their own pace, and prepares students for the future

by giving real-world knowledge and skills (Ustunel & Tokel, 2018), which let

students promptly use their academic capabilities, self-learning skills, and obviously,

computerized know-how in the workplace (Yeou, 2016). The authors also stated that

blended learning positively affects socialization in higher education, increases

student intellectual ability and self-reliance capacity, improves student learning

quality, improves their ability to think critically, and combines advanced technolo-

gies as an operational tool to demonstrate course curriculum to students (Al-shami

et al., 2018). However, prior research authors are mostly recommending blended

learning as an active education in higher education.

Exploring the Role of Mobile Technologies in Higher Education: The... 209

10 Financial Proﬁtability and Complexity Among Learners

Mobile and electronic learning processes have been introduced among learners to

eradicate education barriers. It is undeniable that m-learning reduces the cost of

learners and may bring the whole process to fruition. According to the recent

theoretical developments, this medium of learning has enabled accessibility for

learners. The introduction of mobile learning among people has ensured the sustain-

ability of education. In addition to the beneﬁts of m-learning, it has some deﬁnite

intricacies. This report will ﬁnd the ﬁnancial proﬁtability and complexity with which

learners may deal while obtaining this medium.

10.1 Financial Proﬁtability

The advancement of technology is quickly becoming more efﬁcient and faster.

M-learning has been facilitated by technology that helps enhance the collaboration

between the student and the teacher. Changing the approaches to learning is not only

the motive of this process, but also makes education more affordable for the learners.

In the context of availability, the lectures, tools, and other materials of learning are

available on the required application or website. This helps learners practice any-

time. Mobile technologies have helped to adopt the new learning process that

improves the traditional learning method.

210 S. F. A. Hossain et al.

Indisputably, online learning helps save money and also allows users to access

any content. The books may not be affordable to some people who are from remote

areas of developing or underdeveloped countries. Online courses and classes have

been ﬁnancially convenient for students. The process improves the educational

system while being ﬁnancially beneﬁcial to learners. Through online learning,

learners can get ﬁnancial proﬁts such as saving them money, accommodating in a

comfortable place, commuting costs, expenses of buying materials, and so on. There

is no other alternative way than accessing all the content through m-learning which

diminishes the cost of buying books and other accessories.

The proﬁtability demonstrates that students can learn sustainably. Learners may

collaborate with teachers while connecting online, and it ensures cost-effectiveness.

The cost includes proctoring of exams, which may help invigilators as well as

students save the money of transportation. Online learning such as m-learning and

e-learning does not require learners or trainees to purchase books as all the materials

and PDFs are already uploaded online. Mobile technologies have established mobile

education to enable learning to be affordable and accessible.

Learners can attend classes or courses through mobile learning technologies that

help them learn virtually instead of spending transportation or any other accommo-

dation cost. Because of mobile phone education, people don’t need to leave their city

or areas for training, college, or any other institution. Some people may work while

learning online or reading content by mobile phone which would not hamper the job.

This approach has been inﬂuential because learners can save operational costs and

also printing costs. Previously, they needed to buy printing copies and also print the

documents and ﬁles with the expense. Online courses helped them reduce the cost of

these tools. Online education does not only provide ﬁnancial beneﬁts with education,

but also helps lessen additional costs including meal plans, room-and-board.

10.2 Financial Complexity

Students who live in remote areas cannot afford the Internet and high-end devices.

The cost of mobile devices is a challenging ﬁnancial issue for learners, and the

impact of mobile education from the ﬁnancial perspective may affect the learning

process. Sometimes, online learning requires a high-conﬁgured computer with

available tools that become difﬁcult to obtain.

Exploring the Role of Mobile Technologies in Higher Education: The... 211

11 Conclusion

Learning through mobile education is a progressive way of learning and practicing.

Mobile education technology is a mainstream medium that is helping students with

content, pdfs, and saving time. Analysis of the past decades has shown that online

learning has integrated distance-educated students around the world. The emerging

technology of mobile education has increased education in a great way. The results

have shown that the learners have positive attitudes toward mobile learning and

online education with respect to the current phenomena. Educational technology is

emerging in its learning process. Academic resources are available and accessible to

all learners, making education more ﬂexible. Problems related to the needs of

learners are usually overcome by evaluating their attitudes. Online learning is easy

to adopt and appropriate for exchanging information with faculty and students,

working from anywhere, and also learning new technological features. However,

some learners and teachers have reported issues while working online and using

technology. The technology advancement made the inferiority complex among

learners, and also minimized the social interaction, which makes people antisocial.

Universities, colleges, and other institutions should analyze the effectiveness and

provide proper knowledge in the research and learning process. Mobile education

technologies should be manufactured in a way that students can afford them.

Academicians should formulate a proper policy on the use and operation of mobile

phones in education to avoid misuse and bullying. It is undoubtedly true that

interactivity is the key element of learning and online learning ensures giving prompt

feedback on their performances. Implementing online learning in higher education is

a huge initiative for the future, and this makes education more creative and feasible.

References

Abedin, M. Z., Guotai, C., Moula, F.-E., Azad, A. S. M. S., & Khan, M. S. U. (2019). Topological

applications of multilayer perceptions and support vector machines in ﬁnancial decision support

systems. International Journal of Finance & Economics, 24(1), 474–507. https://doi.org/10.

1002/ijfe.1675

Abedin, M. Z., Moon, M. H., Hassan, M. K., & Hajek, P. (2021). Deep learning-based exchange

rate prediction during the COVID-19 pandemic. Annals of Operations Research, 2021,

1. https://doi.org/10.1007/s10479-021-04420-6

Abedin, M. Z., Guotai, C., Hajek, P., & Zhang, T. (2022). Combining weighted SMOTE with

ensemble learning for the class imbalanced prediction of small business credit risk. Complex &

Intelligent Systems. https://doi.org/10.1007/s40747-021-00614-4

Ahmed, A. A. A., Hassan, I., Pallathadka, H., Keezhatta, M. S., Noorman Haryadi, R., Al

Mashhadani, Z. I., Attwan, L. Y., & Rohi, A. (2022). MALL and EFL learners’speaking:

Impacts of Duolingo and WhatsApp applications on speaking accuracy and ﬂuency. Education

Research International, 2022,1–10. https://doi.org/10.1155/2022/6716474

Ali, W. (2020). Online and remote learning in higher education institutes: A necessity in light of

COVID-19 pandemic. Higher Education Studies, 10(3), 16. https://doi.org/10.5539/hes.

v10n3p16

212 S. F. A. Hossain et al.

Al-shami, S. A., Aziz, H., & Rashid, N. (2018). The adoption of MOOC utilization among

undergraduate students in Universiti Teknikal Malaysia Melaka (UTEM). Journal of Funda-

mental and Applied Sciences, 10(6), 2634–2654. Retrieved from https://www.ajol.info/index.

php/jfas/article/view/172198

Anthony, B., Kamaludin, A., Romli, A., Raffei, A. F., Nincarean, A., Eh Phon, L., Abdullah, D.,

Ming, G. L., Shukor, N. A., Nordin, M. S., & Baba, S. (2019). Exploring the role of blended

learning for teaching and learning effectiveness in institutions of higher learning: An empirical

investigation. Education and Information Technologies, 24(6), 3433–3466. https://doi.org/10.

1007/s10639-019-09941-z

Arnold, D., & Sangrà, A. (2018). Dawn or dusk of the 5th age of research in educational

technology? A literature review on (e-) leadership for technology-enhanced learning in higher

education (2013-2017). International Journal of Educational Technology in Higher Education,

15(1), 24.

Awaji, B., Solaiman, E., & Albshri, A. (2020). Blockchain-based applications in higher education.

In Proceedings of the 5th International Conference on Information and Education Innovations.

https://doi.org/10.1145/3411681.3411688

Azzalis, L., Sato, S., De Mattos, M., Fonseca, F., & Giavarotti, L. (2009). Active learning versus

traditional teaching. Revista de Ensino de Bioquímica, 7(2), 2. https://doi.org/10.16923/reb.

v7i2.186

Baldwin, R. (2021). Technology in education. Retrieved from https://education.stateuniversity.

com/pages/2496/Technology-in-Education-HIGHER- EDUCATION.html

Bao, W. (2020). COVID -19 and online teaching in higher education: A case study of Peking

University. Human Behavior and Emerging Technologies, 2(2), 113–115. https://doi.org/10.

1002/hbe2.191

Bervell, B., & Umar, I. N. (2018). Blended learning or face-to-face? Does tutor anxiety prevent the

adoption of learning management systems for distance education in Ghana? Open Learning:

The Journal of Open, Distance and e-Learning, 35(2), 159–177. https://doi.org/10.1080/

02680513.2018.1548964

Chang-Tik, C. (2018). Impact of learning styles on the community of inquiry presences in multi-

disciplinary blended learning environments. Interactive Learning Environments, 26(6),

827–838. https://doi.org/10.1080/10494820.2017.1419495

Chen, G., Xu, B., Lu, M., & Chen, N. (2018). Exploring blockchain technology and its potential

applications for education. Smart Learning Environments, 5(1). https://doi.org/10.1186/s40561-

017-0050-x

Chen, H., Wu, S., & Shi, J. (2020). Design of online and ofﬂine blending teaching mode. In 2020

IEEE 2nd International Conference on Computer Science and Educational Informatization

(CSEI).https://doi.org/10.1109/csei50228.2020.9142519

Coulouris, G. F., Dollimore, J., & Kindberg, T. (2012). Distributed systems: Concepts and design.

Springer.

Criollo-C, S., Lujan-Mora, S., & Jaramillo-Alcazar, A. (2018). Advantages and disadvantages of

M-learning in current education. In 2018 IEEE World Engineering Education Conference

(EDUNINE).https://doi.org/10.1109/edunine.2018.8450979

Dumford, A. D., & Miller, A. L. (2018). Online learning in higher education: Exploring advantages

and disadvantages for engagement. Journal of Computing in Higher Education, 30(3), 452–465.

https://doi.org/10.1007/s12528-018-9179-z

Edward, C. N., Asirvatham, D., & Johar, M. G. (2018). Effect of blended learning and learners’

characteristics on students’competence: An empirical evidence in learning oriental music.

Education and Information Technologies, 23(6), 2587–2606. https://doi.org/10.1007/s10639-

018-9732-4

Elsisi, M., Tran, M., Mahmoud, K., Mansour, D.-E. A., Lehtonen, M., & Darwish, M. M. F. (2022).

Effective IoT-based deep learning platform for online fault diagnosis of power transformers

against cyberattacks and data uncertainties. Measurement, 190, 110686. https://doi.org/10.1016/

j.measurement.2021.110686

Exploring the Role of Mobile Technologies in Higher Education: The... 213

Elsobeihi, M., & Abu Naser, S. (2022). Effects of mobile technology on human relationships.

Retrieved 13 January 2022, from http://hdl.handle.net/10419/167579

Fabian, K., Topping, K. J., & Barron, I. G. (2015). Mobile technology and mathematics: Effects on

students’attitudes, engagement, and achievement. Journal of Computers in Education, 3(1),

77–104. https://doi.org/10.1007/s40692-015-0048-8

Gao, P., & Lu, M. (2021). Research on closed-loop online and ofﬂine hybrid teaching from the ﬁrst

perspective of students. In 2021 16th International Conference on Computer Science &

Education (ICCSE). doi:https://doi.org/10.1109/iccse51940.2021.9569389

Ghazal, S., Al-Samarraie, H., & Aldowah, H. (2018). “I am still learning”: Modeling LMS critical

success factors for promoting students’experience and satisfaction in a blended learning

environment. IEEE Access, 6, 77179–77201. https://doi.org/10.1109/access.2018.2879677

Guangul, F. M., Suhail, A. H., Khalit, M. I., & Khidhir, B. A. (2020). Challenges of remote

assessment in higher education in the context of COVID-19: A case study of Middle East

college. Educational Assessment, Evaluation and Accountability, 32(4), 519–535. https://doi.

org/10.1007/s11092-020-09340-w

Guernsey, L., Ishmael, K., & Prescott, S. (2020). Online learning in the wake of COVID-19 tips and

resources for PreK-12 with equity in mind, New America. Retrieved from https://www.

newamerica.org/education-policy/edcentral/online-learning-wakecovid-19/

Hadad, S., Meishar-Tal, H., & Blau, I. (2020). The parents’tale: Why parents resist the educational

use of smartphones at schools? Computers & Education, 157, 103984. https://doi.org/10.1016/j.

compedu.2020.103984

Han, S. (2022). Impact of smartphones on students: How age at ﬁrst use and duration of usage affect

learning and academic progress. Technology in Society, 70, 102002. https://doi.org/10.1016/j.

techsoc.2022.102002

Han, I., & Shin, W. S. (2016). The use of a mobile learning management system and academic

achievement of online students. Computers & Education, 102,79–89. https://doi.org/10.1016/j.

compedu.2016.07.003

Hasan, M., Noor, T., Gao, J., Usman, M., & Abedin, M. Z. (2022). Rural consumers’ﬁnancial

literacy and access to FinTech services. Journal of the Knowledge Economy. https://doi.org/10.

1007/s13132-022-00936-9

Hashim, H. (2018). Application of technology in the digital era education. International Journal of

Research in Counseling and Education, 1(2), 1. https://doi.org/10.24036/002za0002

Heﬂin, H., Shewmaker, J., & Nguyen, J. (2017). Impact of mobile technology on student attitudes,

engagement, and learning. Computers & Education, 107,91–99. https://doi.org/10.1016/j.

compedu.2017.01.006

Hossain, S. F. A., Shan, X., & Nurunnabi, M. (2019). Is M-learning a challenge?: Students attitudes

toward the sustainable learning and performance. International Journal of e-Collaboration,

15(1), 21–37. https://doi.org/10.4018/IJeC.2019010102

Hossain, S. F. A., Nurunnabi, M., & Hussain, K. (2021). Continuous mobile devices usage

tendency in the TPACK-based classroom and academic performance of university students.

Technology, Pedagogy and Education, 30(4), 589–607. https://doi.org/10.1080/1475939X.

2021.1933160

Hossain, S. F. A., Xi, Z., Nurunnabi, M., & Anwar, B. (2022). Sustainable academic performance in

higher education: A mixed method approach. Interactive Learning Environments, 30(4),

707–720. https://doi.org/10.1080/10494820.2019.1680392

Hyun, J., Ediger, R., & Lee, D. (2017). Students’satisfaction on their learning process in active

learning and traditional classrooms. International Journal of Teaching and Learning in Higher

Education, 29(1), 108–118. https://doi.org/10.1186/s41239-017-0041-6

Jones, A., Fouladi-Nashta, S., Eady, N., Bedeau, S., & Hafezi, H. (2022). “Just say no”(or at least

ask why) STOMP medication reviews in tower hamlets community learning disability service.

BJPsych Open, 8(S1), S182–S182. https://doi.org/10.1192/bjo.2022.504

Kaplan, A. M., & Haenlein, M. (2010). Users of the world, unite! The challenges and opportunities

of Social Media. Business Horizons, 53(1), 59–68.

214 S. F. A. Hossain et al.

Kearns, L. R. (2012). Student assessment in online learning: Challenges and effective practices.

MERLOT Journal of Online Learning and Teaching, 8(3).

Keengwe, J., & Bhargava, M. (2013). Mobile learning and integration of mobile technologies in

education. Education and Information Technologies, 19(4), 737–746. https://doi.org/10.1007/

s10639-013-9250-3

Kekwaletswe, R. (2007). Social presence awareness for knowledge transformation in a mobile

learning environment. International Journal of Education and Development using ICT, 3(4),

102–109.

Krokos, E., Plaisant, C., & Varshney, A. (2019). Correction to: Virtual memory palaces: Immersion

aids recall. Virtual Reality, 23(1), 17–17. https://doi.org/10.1007/s10055-018-0360-5

Lai, J. W. M., De Nobile, J., Bower, M., & Breyer, Y. (2022). Comprehensive evaluation of the use

of technology in education –Validation with a cohort of global open online learners. Education

and Information Technologies., 27, 9877. https://doi.org/10.1007/s10639-022-10986-w

Leinonen, R., Sugawara, H., Shumway, M., & International Nucleotide Sequence Database Col-

laboration. (2010). The sequence read archive. Nucleic Acids Research, 39(suppl_1), D19–D21.

Liguori, E., & Winkler, C. (2020). From ofﬂine to online: Challenges and opportunities for

entrepreneurship education following the COVID-19 pandemic. Entrepreneurship Education

and Pedagogy, 3(4), 346–351. https://doi.org/10.1177/2515127420916738

Liu, Z., & Han, Z. (2020). Exploring trends of potential user experience of online classroom on

virtual platform for higher education during COVID-19 epidemic: A case in China. In 2020

IEEE International Conference on Teaching, Assessment, and Learning for Engineering

(TALE).https://doi.org/10.1109/tale48869.2020.9368349

Long, H. M., & Bouck, E. C. (2022). Calculators and online games: Supporting students with

learning disabilities in mathematics. Intervention in School and Clinic, 2022,

105345122210937. https://doi.org/10.1177/10534512221093787

Massey, C. C., Dogan, S., Muhammad, E., Hogan, E., Head, C., & Kim, J. (2022). The effect of

ﬂipped instruction on special education preservice teachers’perceptions. International Journal

for the Scholarship of Teaching and Learning, 16(2), 10.20429/ijsotl.2022.160213.

Mindster. (2020, December 25). 10 best free educational apps for students & kids learning.

Retrieved from https://mindster.com/free-educational-appsstudents/#Best_10_Free_Educa

tional_Apps_for_Students

Moslehi, P., Rilling, J., & Adams, B. (2022). A user survey on the adoption of crowd-based

software engineering instructional screencasts by the new generation of software developers.

Journal of Systems and Software, 185, 111144. https://doi.org/10.1016/j.jss.2021.111144

Nguyen, D. C., Ding, M., Pathirana, P. N., Seneviratne, A., Li, J., Niyato, D., & Poor, H. V. (2022).

6G Internet of Things: A comprehensive survey. IEEE Internet of Things Journal, 9(1),

359–383. https://doi.org/10.1109/JIOT.2021.3103320

Paul, J., & Jefferson, F. (2019). A comparative analysis of student performance in an

online vs. face- to-face environmental science course from 2009 to 2016 (p. 1). Frontiers in

Computer Science. https://doi.org/10.3389/fcomp.2019.00007

Radianti, J., Majchrzak, T. A., Fromm, J., & Wohlgenannt, I. (2020). A systematic review of

immersive virtual reality applications for higher education: Design elements, lessons learned,

and research agenda. Computers & Education, 147, 103778. https://doi.org/10.1016/j.compedu.

2019.103778

Raja, R., & Nagasubramani, P. C. (2018). Impact of modern technology in education. Journal of

Applied and Advanced Research, S33–S35. https://doi.org/10.21839/jaar.2018.v3is1.165

Razeeth, M. S. S., Kariapper, R. K. A. R., Pirapuraj, P., Nafrees, A. C. M., Rishan, U. M., &

Nusrath Ali, S. (2019). E-learning at home vs traditional learning among higher education

students: A survey- based analysis. In 9th International Symposium 2019. Retrieved from http://

ir.lib.seu.ac.lk/handle/123456789/3909

Richardson, J. C., & Swan, K. (2003). Examining social presence in online courses in relation to

student’s perceived learning and satisfaction. Journal of Asynchronous Learning Networks, 7,

68–88.

Exploring the Role of Mobile Technologies in Higher Education: The... 215

Rmenardi. (2012). More smart phones than desktop computers? It’stime. Retrieved from http://

torsionmobile.com/2012/01/26/more-smart-phones-than-desktop-computers-its-time/

Rogers, S. (2019). Virtual reality: The learning aid of the 21st century. Retrieved from https://www.

forbes.com/sites/solrogers/2019/03/15/virtual-reality-the-learning-aid-of-the-21st-century/?

sh=39d4bda0139b

Rowe, N. C. (2004). Cheating in online student assessment: Beyond plagiarism. Online Journal of

Distance Learning Administration, 7(2), 1–10.

Sattarov, A., & Khaitova, N. (2019). Mobile learning as new forms and methods of increasing the

effectiveness of education. European Journal of Research and Reﬂection in Educational

Sciences, 7(12), 1169–1175.

Schindler, L. A., Burkholder, G. J., Morad, O. A., & Marsh, C. (2017). Computer-based technology

and student engagement: A critical review of the literature. International Journal of Educational

Technology in Higher Education, 14(1), 1–28. https://doi.org/10.1186/s41239-017-0063-0

Scholz, K. W., Komornicka, J. N., & Moore, A. (2021). Gamifying history: Designing and

implementing a game-based learning course design framework. Teaching & Learning Inquiry,

9(1), 99–116. https://doi.org/10.20343/teachlearninqu.9.1.9

Senior, J. (2022). Vocabulary taught via Mobile application gamiﬁcation: Receptive, productive

and Long-term usability of words taught using Quizlet and Quizlet live. In 2022 International

Conference on Business Analytics for Technology and Security (ICBATS) (pp. 1–7). IEEE.

https://doi.org/10.1109/ICBATS54253.2022.9759019

Shahzad, A., Hassan, R., Aremu, A. Y., Hussain, A., & Lodhi, R. N. (2020). Effects of COVID-19

in E-learning on higher education institution students: The group comparison between male and

female. Quality & Quantity, 55(3), 805–826. https://doi.org/10.1007/s11135-020-01028-z

Sharples, M. (2007). Big issues in Mobile learning: Report of a workshop by the kaleidoscope

network of excellence mobile learning initiative. Learning Sciences Research Institute, Univer-

sity of Nottingham.

Shi, Y., & Lin, X. (2021). Exploring the characteristics of adults’online learning activities: A case

study of EdX online institute. Research in Learning Technology, 29,1–13. https://doi.org/10.

25304/rlt.v29.2622

Shyshkanova, G., Zaytseva, T., & Frydman, O. (2017). Mobile technologies make education a part

of everyday life. Information and Learning Science, 118(11/12), 570–582. https://doi.org/10.

1108/ils-03-2017-0019

Subhash, S., & Cudney, E. A. (2018). Gamiﬁed learning in higher education: A systematic review

of the literature. Computers in Human Behavior, 87, 192–206.

Subramaniam, S. R., & Muniandy, B. (2019). The effect of ﬂipped classroom on students’

engagement. Technology, Knowledge and Learning, 24(3), 355–372. https://doi.org/10.1007/

s10758-017-9343-y

Sun, Z., & Qiu, X. (2017). Developing a blended learning model in an EFL class. International

Journal of Continuing Engineering Education and Life-Long Learning, 27(1/2), 4. https://doi.

org/10.1504/ijceell.2017.080998

Tarteer, S., Badah, A., & Khlaif, Z. N. (2021). Employing Google classroom to teach female

students during the COVID-19 pandemic. Computers in the Schools, 38(4), 300–321. https://

doi.org/10.1080/07380569.2021.1988318

Traxler, J. (2005). Deﬁning mobile learning. IADIS International Conference Mobile Learning.

Retrieved from http://iadis.net/dl/ﬁnal_uploads/200506C018.pdf

Troussas, C., Krouska, A., & Sgouropoulou, C. (2020). Collaboration and fuzzy-modeled person-

alization for mobile game-based learning in higher education. Computers & Education, 144,

103698. https://doi.org/10.1016/j.compedu.2019.103698

Turkanovic, M., Holbl, M., Kosic, K., Hericko, M., & Kamisalic, A. (2018). EduCTX: A

blockchain-based higher education credit platform. IEEE Access, 6, 5112–5127. https://doi.

org/10.1109/access.2018.2789929

216 S. F. A. Hossain et al.

Ustunel, H. H., & Tokel, S. T. (2018). Distributed scaffolding: Synergy in technology-enhanced

learning environments. Technology, Knowledge and Learning, 23(1), 129–160. https://doi.org/

10.1007/s10758-017-9299-y

Van Laer, S., & Elen, J. (2020). Adults’self-regulatory behaviour proﬁles in blended learning

environments and their implications for design. Technology, Knowledge and Learning, 25(3),

509–539. https://doi.org/10.1007/s10758-017-9351-y

Vijayakumar, S. (2022). Foreign language learning then, now and after COVID-19: An exploration

of digital tools to augment the receptive and productive skills of language learners. In

A. Hamdan, A. E. Hassanien, T. Mescon, & B. Alareeni (Eds.), Technologies, artiﬁcial

intelligence and the future of learning post-COVID-19 (pp. 283–302). Springer. https://doi.

org/10.1007/978-3-030-93921-2_17

Vujić, A., & Szabo, A. (2022). Hedonic use, stress, and life satisfaction as predictors of smartphone

addiction. Addictive Behaviors Reports, 15, 100411. https://doi.org/10.1016/j.abrep.2022.

100411

Yang, F., Qiao, Y., Abedin, M. Z., & Huang, C. (2022). Privacy-preserved credit data sharing

integrating blockchain and federated learning for industrial 4.0. IEEE Transactions on Indus-

trial Informatics,1–1. https://doi.org/10.1109/TII.2022.3151917

Yeou, M. (2016). An investigation of students’acceptance of Moodle in a blended learning setting

using technology acceptance model. Journal of Educational Technology Systems, 44(3),

300–318. https://doi.org/10.1177/0047239515618464

Yip, K. H., Lo, P., Ho, K. K., & Chiu, D. K. (2020). Adoption of mobile library apps as learning

tools in higher education: A tale between Hong Kong and Japan. Online Information Review,

45(2), 389–405. https://doi.org/10.1108/oir-07-2020-0287

217

Knowledge Mining from Health Data:

Application of Feature Selection

Approaches

Md. Rabiul Auwul, Md. Ajijul Hakim, Fahmida Tasnim Dhonno,

Nusrat Afrin Shilpa, and Mohammad Zoynul Abedin

Abstract This paper aims to measure the performance of feature selection

approaches for mining knowledge from health datasets. We compare seven popular

knowledge mining approaches, namely relaxed Lasso, random forest, ReliefF,

OneR, information gain, T-test, and Chi-squared test. The support vector machine

(SVM) classiﬁer applies to determine the accuracy and area under the curve (AUC)

values of the knowledge miners. We use six popular Affymetrix and cDNA datasets.

The results reveal that the relaxed lasso works well with Affymetrix, and the relaxed

Lasso with random forest approaches perform well with the cDNA datasets. This

paper will enrich the existing literature and assist to identify the best feature for

knowledge mining in the health informatics domain.

Keywords Knowledge mining · Feature selection · Classiﬁcation · Cancer data ·

SVM · Affymetrix · cDNA datasets

M. R. Auwul

Department of Mathematics, Faculty of Science and Technology, American International

University-Bangladesh, Dhaka, Bangladesh

M. A. Hakim

Foreign Exchange and Remittance Department, Travelex Qatar, Golbex Business Center, Doha,

Qatar

F. T. Dhonno · N. A. Shilpa

Department of Finance and Banking, Hajee Mohammad Danesh Science and Technology

University, Dinajpur, Bangladesh

M. Z. Abedin (✉)

Department of Finance, Performance and Marketing, Teesside University International Business

School, Teesside University, Middlesbrough, Tees Valley, UK

e-mail: m.abedin@tees.ac.uk

M. Z. Abedin, P. Hajek (eds.), Novel Financial Applications of Machine Learning

and Deep Learning, International Series in Operations Research & Management

Science 336, https://doi.org/10.1007/978-3-031-18552-6_13

218 M. R. Auwul et al.

1 Introduction

Knowledge mining (data mining) is the investigation and exploration of a large

amount of data to ﬁnd signiﬁcant patterns and directions. Extracting knowledge from

the existing information is the principal job of knowledge mining. Knowledge

mining depends on two parameters; one of them is the association rule that produced

by scrutinizing data for regular patterns, and then discovering the most signiﬁcant

associations within the data by using the support and conﬁdence criteria. The second

parameter includes Classiﬁcation, Sequence or Path Analysis, Clustering, and Fore-

casting. In the modern era, the rapid growth of data in every ﬁeld is quickly

mounting with time, as is the importance of knowledge mining (Lu et al., 2022).

The health sector is one of the fast developing and challenging sections in data

mining, and it is becoming popular day by day. All the parties involved in health

section are greatly beneﬁted through data mining like the healthcare insurers identify

fraud and misuse; healthcare groups create client relationship management deci-

sions, and patients get improved and more affordable healthcare facilities (Koh

&Tan, 2005). The high-dimensional microarray data used in health research are

mostly constructed from two vital chips: Affymetrix and cDNA that are too complex

and big to be handled and investigated by classical approaches. Knowledge mining

offers the procedure and equipment to make over these large volumes of data into

beneﬁcial material for decision making.

These microarray data formed as an array with relevant and redundant features

and samples. Feature selection is an important part to select a subset of relevant

features to build effective prediction models, especially for classiﬁcation purposes.

To select the signiﬁcant features from high-dimensional data, there exist several

works in literature. But no one can exactly show the application of feature selection

methods differentially in the ﬁelds of Affymetrix and cDNA microarray data

(Rahman et al., 2021).

Feature selection is more essential for high-dimensional data to improve the

performance of prediction models by eliminating irrelevant and redundant features

in cancer research mostly in gene expressed data that are provided mostly from DNA

microarray technology. The problems come in front of researchers when these high-

dimensional gene expressed data contain huge number of genes but a few number of

samples. The importance of feature selection helps to remove irrelevant genes in

high-dimensional data with small sample to prevent declining the classiﬁcation

accuracy from the inﬂuence of redundant genes. In this paper, we explore the

seven most important feature selection approaches, namely, Relaxed Lasso, Random

Forest, ReliefF, OneR, Information Gain, T-test, and Chi-squared test on six popular

Affymetrix and cDNA cancer gene expressed datasets.

The results obtained from the analysis of seven popular feature selection

approaches reveal that Relaxed Lasso works well with Affymetrix and Relaxed

Lasso and Random Forest methods work well with cDNA datasets compared to

other methods. This study provides outlines of applied assessment to access the

results of feature selection in gene expression cancer datasets.

Knowledge Mining from Health Data: Application of Feature Selection Approaches 219

In the healthcare section through the ﬁndings of our paper, feature selection

methods will be more effective in areas such as predictive medicine, recognition

of fraud and misapplication, consumer relationship administration, controlling of

healthcare and measuring the effectiveness of deﬁnite treatments, as well as used to

reduce costs by growing efﬁciencies, progress patient superiority of life, and possi-

bly, utmost notably, protect the lives of more patients. For academia, researchers can

easily ﬁnd the best feature selection approaches for Affymetrix and cDNA data when

they work with knowledge mining such as clustering, classiﬁcation, etc., and will

contribute to the health section.

The rest of this paper is organized as follows. Section 2delivers a brief review of

the literature. Section 3presents a brief description of the methods and feature

selection approaches used in this paper. Section 4describes the results and discus-

sion. Finally, Sect. 5concludes the paper.

2 Related Works

To reduce the dimensionality and select relevant genes, various features/gene selec-

tion approaches existed. Filters, wrappers, and embedded methods are three catego-

ries of feature selection approaches. The important features are selected by

measuring the correlation between individual features and output class labels,

without involving any learning algorithm through ﬁlter methods such as ReliefF

(Kira & Rendell, 1992), Information Gain (Dagliyan et al., 2011),T-test (Abedin

et al., 2018), and Chi-squared test (Guotai et al., 2017). Through wrapper methods, a

subset of features evaluated by machine learning algorithm employs a search

approach to look through the space of potential feature subsets, evaluating each

subset based on the quality of the performance of a given algorithm. The sequential

feature selection method such as forward/backward selection is an example of

wrapper method that is also known as greedy method for its searching strategy.

Wrapper methods are more complex and expensive than simpler ﬁlter methods.

Through embedded methods a penalty term is added against complexity to reduce

the degree of over ﬁtting or variance of a model by adding more bias such as L1

(or Lasso) regression for generalized linear regression (Tibshirani, 1996), relaxed

lasso (Abedin et al., 2019). The embedded methods are usually faster than the

wrapper methods and able to provide a suitable feature subset for the learning

algorithm.

The correlation-based feature selection approach is used by Harb and Desuky to

develop the classiﬁcation of health datasets (Harb & Desuky, 2014). Jovićet al.

(2015) reviewed several ﬁlter, wrapper, and embedded feature selection methods

with their application. They showed the best for text mining, image processing,

computer vision, and industrial application. The Lasso feature selection approach

with information gain has been compared by Kamkar et al. to build clinical

prediction models (Guo et al., 2015). Lasso and ridge regression are being compared

by Fonti and Belitser to implement feature selection on high-dimensional datasets

(Fonti & Belitser, 2017). B. Remeseiro and V. Bolon-Canedo (2019) reviewed six

state-of-the-art algorithms: CFS, INTERACT, InfoGain, CFS, ReliefF, and

SVM-RFE for medical application in terms of four classiﬁcation algorithms, namely:

Naive Bayes, SVM, C4.5, and K-NN. They showed that the classiﬁer performance

improved with signiﬁcant selected features. Chuanze Kang et al. (2019)showed the

effect of feature gene selection ReliefF, Relaxed Lasso, Information gain, and

Kruskal–Wallis rank sum test for eight microarray data with several classiﬁers.

Relaxed Lasso gave better results for all microarray datasets. ShrutiKaushik et al.

compared the traditional feature selection approaches on a healthcare dataset for

classiﬁcation purposes involving several attributes (Kaushik et al., 2019).

220 M. R. Auwul et al.

The above literature has shown the application of feature selection approaches on

healthcare data, but no one has analyzed them on Affymetrix and cDNA microarray

data. In this paper, we compare seven popular feature selection approaches, namely:

Relaxed Lasso, Random Forest, ReliefF, OneR, Information Gain, T-test, and

Chi-squared test on six popular Affymetrix and cDNA datasets.

3 Material and Methods

3.1 Datasets

In this paper, we used three Affymetrix and three cDNA datasets to evaluate the

performance of feature selection approaches. These datasets have been used in many

other research papers, among which we will mention only a few examples. Datasets

of CNS, Lung DLBCL have been used to analyze the impact of selecting signiﬁcant

features on the classiﬁcation performance by Chuanze Kang et al. (2019). A com-

parative study of clustering algorithms for several cancer gene expression data like

Shipp, Alizada, Bittner, and Chen datasets is used by Marcilio CP de Souto et al.

(2008) (Table 1).

Table 1 Affymetrix and cDNA datasets used in this paper

Dataset Chip #Sample Dist. Classes #Genes

CNS (Pomeroy et al., 2002) Affy 60 21,39 7129

Lung (Beer et al., 2002) Affy 86 62,24 7129

DLBCL/Shipp (Shipp et al., 2002) Affy 77 58,19 7129

Alizadeh-V1 (Alizadeh et al., 2000) cDNA 42 21,21 4022

Bittner (Bittner et al., 2000) cDNA 38 19,19 8067

Chen (Chen et al., 2002) cDNA 180 104,76 22,699

Knowledge Mining from Health Data: Application of Feature Selection Approaches 221

3.2 Feature Selection Approaches

Relaxed Lasso

A generalization method proposed by Meinshausen (2006) as of soft-thresholding

and hard-thresholding known as relaxed Lasso is deﬁned as:

λ,φ=argmin

n-1XXi-YT

iβ:1ρλ



2þφλ β

1,ð1Þ

for λ2[0, 1) and φ2(0, 1]. The indicator functions on the set of variables

⊆{1, ...,p} noted as 1ρ

,8k2{1, ...., p}:

β:1ρλ=0, k=2ρλ

βk,k2ρλ

:ð2Þ

The predictor variables in the set ρ

are measured for the relaxed Lasso estimator.

For the variable selection part, the parameter λcontrols in ordinary Lasso estimation.

The shrinkage of the coefﬁcients is controlled by the relaxation parameter φ. For

example, for φ=1, the relaxed Lasso estimators tend to Lasso estimators. For φ<

1, the shrinkage of relaxed Lasso is reduced parallel to ordinary Lasso estimation.

The above deﬁnition would produce a decadent solution in the case of φ=0.

Accordingly, it minizes the limitation of the relaxed Lasso for φ=0 of the above

deﬁnition for φ→0. All the coefﬁcients in the model ρ

are estimated by the

OLS-solution.

Step 1: Compute all ordinary Lasso solutions, e.g., with the Lars-algorithm in

Efron et al. (2004) under the Lasso modiﬁcation. Let ρ

,...,ρ

be the resulting set of

s models. Let λ

>...>λ

=0 be a sequence of penalty terms so that ρ

=ρ

iff, λ2

(λ

,λ

k-1

Step 2: Let gkðÞ=β

λk-β

λk-1=λk-1-λk

ðÞfor each k=1, ...,c. Through this

direction, ordinary Lasso solut ions can be estimated. Let β

=β

λkþλkgkðÞ. If there is

at least one component lso that signβ

l≠signβ

λk

l, then relaxed Lasso solutions

for λ2Λ

have to be computed as in Step 2 of the simple algorithm. Otherwise, all

relaxed Lasso solutions for λ2Λ

and φ2[0, 1] are given by linear interpolation

between β

λk-1.

Let Y~N(0, Σ), then the response variable can be written by the following linear

combination:

X=YTβþε,ð3Þ

where ε~N(0, σ

), the loss function of relaxed Lasso under parameter λand φis

deﬁned as:

 

222 M. R. Auwul et al.

Lλ;φðÞ=E X -YTβ

λ,φ2-σ2:ð4Þ

For sporadic high-dimensional data, a relaxed Lasso is more appropriate.

Random Forest

Random forest (RF) is an embedded feature selection approach proposed by

Breiman (2001) that generates numerous decision trees based on averaging random

selection of response variables of training set. The importance of a variable in a data

set Z

={(a

)}, j=1, 2, , ...nis measured by ﬁtting a random forest to the data

and the error for each data point is calculated and averaged over the forest. The

importance score for the j

feature is computed by averaging the difference in error

before and after the permutation for all the trees. Select those features that produce

larger values for this score.

ReliefF

An extension version of Relief (Kira & Rendell, 1992)that randomly procures a

sample S each time from training samples is known as ReliefF (Robnik-Sikonja &

Kononenko, 2003). The weight values are computed and updated by ﬁndings k

nearest neighbor samples from samples of the same class as Sand samples of

different class from S, respectively, as follows:

WZ=WZ-Xk

i=1diff Z,S,HðÞ=nk

þXB=2class SðÞ PBðÞ

P class SðÞð×Xk

i=1diff Z,S,NiBðÞÞð=nk:

hð5Þ

The ith nearest neighbor sample in class Bis denoted as N

(B) and diff(g,t

)

denotes the difference between sample t

and sample t

in the feature g. The formula

for diff(g,t

)ifgis discrete is the following:

diff g,t1,t2

ðÞ=0, t1g½=t2g½

1, t1g½≠t2g½

:ð6Þ

The formula for diff(g,t

)ifgis continuous is:

diff g,t1,t2

ðÞ=t1g½-t2g½jj

max gðÞ-min gðÞ

:ð7Þ

The feature with high correlation with the class gives the highest weight, and the

features are selected according to the orderly weights (Kang et al., 2019).

Information Gain

An entropy-based feature selection method computes the mutual information for

each attribute and class and then yields an ordered ranking of all of the features

known by information gain (IG). If Xand Yare the features and p(x) is the marginal

probability density function, then the entropy of given dataset is equated as:

ﬃﬃﬃﬃﬃ

Knowledge Mining from Health Data: Application of Feature Selection Approaches 223

HXðÞ=-

Xx2XpxðÞlog 2pxðÞ½:ð8Þ

The conditional entropy of Xis given that Yis observed before with the condi-

tional probability p(x|y),

jYðÞ=-

Xx2XpxðÞ

Xx2Xpx

jyðÞlog 2px

jyðÞ½:ð9Þ

Finally, the information gain metric is:

IG =HXðÞ-HX

jYðÞ:ð10Þ

Features are ranked according to the IG value. Whose IG value is greater are more

important features than others (Dagliyan et al., 2011).

OneR

Rule-based embedded feature selection methods construct one rule in training data

for each attribute and select rule with smallest error and so that the accuracy could be

optimized (Holte, 1993). The features are selected according to the ordered accuracy

to the corresponding rules. It follows a decision tree approach. For example, if

R=(x,y) is a classiﬁcation rule with precondition xthat executes a sequence of

tests that can be estimated as true or false and yis a class that can be suitable to

occurrences enclosed by rule R. For OneR, a one-level decision tree constructs and

tests an individual attribute at a time and branches for every value of that attribute.

T-Test

To test the independence of two features, the T-test proposed by Gosset is used to

quantify the signiﬁcance of each single feature by determining the following t-

statistic with respect to the class:



y1-

sp2

p,ð11Þ

where sp=ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

y1þs2



rfor n=n

,s2

y1and s2

y2are the unbiased estimators of

the variances of the two samples. The p-value based on these tscores then computed,

and based on these p-values (the smaller the p-value, the more important the feature),

the important features are selected.

Chi-Squared Test

To test the independence of two features, Chi-squared (χ

) is used that quantiﬁes the

signiﬁcance of each single feature by determining the following Chi-squared statistic

with respect to the class:

224 M. R. Auwul et al.

χ2

d=XObs -ExpðÞ

Exp ,ð12Þ

where Obs are the observed values, Exp are the expected values, and dare the

degrees of freedom. The aim of every feature selection method is to select those

features that are highly dependent on the response. The larger the Chi-squared value

means that the observed values are close to the expected values, the higher the

importance of that feature. This method gives misleading results for small frequen-

cies (especially <5).

Classiﬁer Application

Classiﬁcation is a popular data mining process for classifying test data based on

training data. For ﬁnding the accuracy of feature selection methods, we applied an

SVM classiﬁer (Boser et al. 1992) with ten-fold cross-validation. The standardiza-

tion of each feature was also used, which reduces the learning time and equalizes the

impact of each predictor on the target variable. SVM is used to ﬁnd the hyperplane

that separates two different sets of features with the maximum distance of the

hyperplane to the nearest feature from both sets.

The linear SVM formula is as follows:

S=

w:

y-b:ð13Þ

Here, for the hyperplane, yis the input vector and wis the normal vector with the

following distance:

d=1

Wkk

2:ð14Þ

If y

is the jth training sample and z

is the correct output of the SVM for the jth

training sample, then the maximum distance dcan be expressed as:

min

w,b

subject to zjwyj-b



≥1:ð15Þ

For the positive and negative samples, z

is +1 and -1, respectively.

Performance Analysis

In order to assess the performance of different feature selection methods, we

calculate the area under the receiving operating characteristics curve (AUC) and

accuracy of each method, Accuracy =(TP + TN)/(TP + FP + TN + FN), where TP,

TN, FP, and FN denote the number of true positive, number of true negative, number

of false positive, and number of false negative, respectively. Based on these two

parameters, we declare a method as a good performer if it produces larger values of

Accuracy and AUC values (Fig. 1).

Knowledge Mining from Health Data: Application of Feature Selection Approaches 225

Apply different FS methods and rank fea-

tures/genes according to its statistic

Apply 10-fold cross validation

methods with SVM classifier

Select first 2, 3,….., 30 features respectively

from ranked features for different FS methods

Calculate Accuracies

and AUC values

Cancer Gene Ex-

pressed Data

Data Preprocessing

and Normalizing

Fig. 1 Flowchart of the methodology of this paper

4 Results and Discussion

Chuanze Kang et al. (2019) speciﬁed that the classiﬁcation accuracy reaches the

highest value when the number of feature genes is below 30 for all datasets. Hence,

the performance of feature selection methods compared in this paper within the

domain of 2 to 30 numbers of feature.

Figure 2shows the association between the number of feature genes (NF) in the

range of 2–30 and the classiﬁcation accuracy (ACC) for three Affymetrix data.

Figure 2a shows the performance of CNS data, Fig. 2b for Lung data, and Fig. 2c

for Shipp data. For the Shipp dataset and the CNS dataset, the accuracies touch

almost 100%, whereas other approaches have more variations in accuracies for 2–30

features, and there is no development with the increasing of NF. Figure 2b shows

that the Relaxed Lasso has the highest accuracy for Lung datasets. When NF is larger

than 13, the features selected by the other three methods do not hold the resultant in

variations of accuracy, except the T-test and random forest with the classiﬁcation.

The other methods probably select redundant genes foremost to decrease the accu-

racy with increasing the NF. This ﬁgure shows that the Relaxed Lasso has the

highest accuracies. Hence, we may conclude that Relaxed Lasso achieves better

and is more appropriate for feature selection of high-dimensional and small-sample

Affymetrix data.

Figure 3shows the association between the number of feature genes (NF) in the

range of 2–30 and the classiﬁcation accuracy (ACC) for three cDNA data. Figure 3a

shows the performance for Bittner data, Fig. 3b is for Alizada data, and Fig. 3c is for

Chen data. Figure 3shows that the Relaxed Lasso has the highest accuracy. For the

Bittner dataset and the Alizada dataset, the ACC is almost 100% for Relaxed Lasso.

Whereas other methods have more variations in accuracies, there has been no

development with increasing of NF. For Chen datasets, Relaxed Lasso has the

highest accuracy for more than 13 NF and for less than 13 NF; Random Forest

gives a better accuracy than Relaxed Lasso and the others. The feature selected by

226 M. R. Auwul et al.

100

2 5 8 11141720232629

Accuracy

Number of Selected Features

(a) CNS Data

100

25811141720232629

Accuracy

Number of Selected Features

(b) Lung Data

100

2 5 8 11141720232629

Accuracy

Number of Selected Features

Relaxe lasso

Random forest

ReliefF

Information Gain

OneR

t-test

Chi-Squre test

Fig. 2 Accuracy plot for Affymetrix data

methods does not uphold the resultant in variations of accuracy except for Relaxed

Lasso and Random Forest with the classiﬁcation. The other methods probably select

redundant genes foremost to decrease accuracy with the increase in NF. Hence,

Knowledge Mining from Health Data: Application of Feature Selection Approaches 227

100

2 5 8 11 14 17 20 23 26 29

Accuracy

Number of Selected Features

(a) Bittner Data

100

2 5 8 11 14 17 20 23 26 29

Accuracy

Number of Selected Features

(b) Alizada Data

100

2 5 8 11141720232629

Accuracy

Number of Selected Features

Relaxed Lasso

Random Forest

ReliefF

Informaon Gain

OneR

t-test

Chi-Squre test

Fig. 3 Accuracy plot for three cDNA data

Feature Selection Methods

Relaxed Lasso and Random Forest achieve better results and are more appropriate

for feature selection of high-dimensional and small-sample cDNA data. Table 2

shows the average AUC values of the feature selection methods. The relaxed lasso

gives the maximum values of AUC on an average: for CNS, Lung, Ship, Alizada-V1

andspiepr and Fig. 3(b) is for Alizada data, and Fig. 3c is for Chen data. Table 3

228 M. R. Auwul et al.

Table 2 Performance evaluation of the average AUC values of feature selection methods

Affymetrix Datasets cDNA Datasets

CNS Lung Shipp Alizada-V1 Bittner Chen

Relaxed Lasso 0.844 0.845 0.962 0.981 0.978 0.939

Random Forest 0.733 0.703 0.929 0.970 0.951 0.945

ReliefF 0.625 0.531 0.874 0.910 0.916 0.904

Information gain 0.596 0.527 0.890 0.903 0.900 0.919

OneR 0.590 0.532 0.878 0.851 0.894 0.923

T-test 0.795 0.847 0.923 0.958 0.945 0.928

Chi-Square test 0.586 0.538 0.870 0.879 0.812 0.944

Bold values indicate the maximum AUC across the datasets and feature selection methods

Table 3 Efﬁciency measurement of the feature selection approaches by the Mann-Whitney U test

Datasets Random Forest ReliefF IG OneR T-test Chi-squared test

629 817.5 841 841 578.5 839.5

CNS 0.00115 6.03E-

5.40E-

4.90E-

0.014 6.28E-11

Yes Yes Yes Yes Yes Yes

777 841 841 841 592.5 840

Lung 3.06E-08 6.20E-

5.69E-

5.33E-

0.008 5.26E-11

Yes Yes Yes Yes Yes Yes

753 823 779.5 786.5 823 786.5

Shipp 6.53E-08 8.16E-

6.81E-

3.38E-

5.51E-

3.56E-09

Yes Yes Yes Yes Yes Yes

577.5 795.5 783 822 673 813

Alizada-

0.00875 3.19E-

1.03E-

2.52E-

3.17E-

6.42E-10

Yes Yes Yes Yes Yes Yes

681 763 771.5 772.5 698 820

Bittner 2.23E-05 5.09E-

2.44E-

2.41E-

8.72E-

2.49E-10

Yes Yes Yes Yes Yes Yes

297 700 464.5 446 647.5 285.5

Chen 0.05517 1.40E-

0.498 0.697 4E-04 0.03629

No Yes No No Yes Yes

N.B: The ﬁrst, second, and third rows of each dataset are the Mann-Whitney U test score, p-Values,

and the statement on the average efﬁciency of Relaxed Lasso greater (Yes) or not (No)

shows the Mann-Whitney U test score and their corresponding p-values for six

datasets. The results indicate that the Relaxed Lasso feature selection methods are

more efﬁcient than the other six algorithms for the CNS, Lung, Shipp, alizada-V1,

and Bittner datasets and for the Chen dataset Relaxed Lasso performed better than

ReliefF and T-test. The efﬁciency of the Random Forest methods is comparatively

higher than that of the others except for Relaxed Lasso for the ﬁrst ﬁve datasets, and

for Chen datasets its performance is better than Relaxed Lasso.

Knowledge Mining from Health Data: Application of Feature Selection Approaches 229

5 Conclusion

To investigate and explore a large amount of existing information, knowledge

mining plays a signiﬁcant role in the health sector. The ﬁndings indicate that

knowledge mining is an important and prerequisite part for the stakeholders such

as cancer biomarker, genetic pattern for infectious diseases, medicine analytics, and

so on.

The superior nature of microarray data is the huge number of genes but small

number of samples that generates the prerequisite for important gene selection. To

classify large volumes of data, feature selection is a vital issue. There are abundant

studies on feature selection to identify cancer classiﬁcation using microarray gene

expression data. But none of these papers include the performance of feature

selection approaches in different sections for Affymetrix and cDNA microarray

datasets.

This paper has reviewed and analyzed seven popular feature selection

approaches, namely: Relaxed Lasso, Random Forest, ReliefF, OneR, Information

Gain, T-test, and Chi-squared test for cancer classiﬁcation. A widespread analysis

has been conducted and compared these feature selection approaches separately

across six Affymetrix and cDNA datasets. The performance evaluation is conducted

by ﬁnding their accuracy and AUC values with SVM classiﬁer. From our investi-

gation we found that Relaxed Lasso works well with Affymetrix, and Relaxed Lasso

and Random Forest approaches work well with cDNA datasets comparatively with

other approaches.

Through the ﬁndings of our paper in healthcare sector, feature selection

approaches will be more effective in areas such as ﬁnding biomarker cancer gene,

predictive medicine for infectious diseases such as COVID-19, reduction of medical

costs by increasing the efﬁciency of methods, progressing patient superiority of life,

and possibly most importantly, protecting the lives of more patients by using

clustering, classiﬁcation, pattern recognition, and other knowledge mining

approaches. In the academia sector, researchers can easily ﬁnd the best feature

selection approaches for Affymetrix and cDNA data when they work with knowl-

edge mining approaches and will contribute to the health section.

230 M. R. Auwul et al.

Regarding future research, we will explore the performance of these feature

selection approaches with big data in deep learning. This will be more reliable,

informative, and enrich the existing literature.

References

Abedin, M. Z., Guotai, C., Colombage, S., & Moula, F. E. (2018). Credit default prediction by

using a support vector machine and a probabilistic neural network. Journal of Credit Risk, 14(2),

1–27.

Abedin, M. Z., Guotai, C., Moula, F. E., Zhang, T., & Hassan, M. K. (2019). An optimized support

vector machine intelligent technique using optimized feature selection methods: Evidence from

Chinese credit approval data. Journal of Risk Model Validation, 13(2), 1–46.

Alizadeh, A. A., Eisen, M. B., Davis, R. E., Ma, C., Lossos, I. S., Rosenwald, A., et al. (2000).

Distinct types of diffuse large B-cell lymphoma identiﬁed by gene expression proﬁling. Nature,

403, 503–511.

Beer, D. G., Kardia, S. L., Huang, C. C., Giordano, T. J., Levin, A. M., Misek, D. E., Lizyness,

M. L., Kuick, R., Hayasaka, S., Taylor, J. M., Iannettoni, M. D., Orringer, M. B., & Hanash,

S. (2002). Gene-expression proﬁles predict survival of patients with lung adenocarcinoma.

Nature Medicine, 8, 816–824.

Bittner, M., Meltzer, P., Chen, Y., Jiang, Y., Seftor, E., Hendrix, M., et al. (2000). Molecular

classiﬁcation of cutaneous malignant melanoma by gene expression proﬁling. Nature,

406(6795), 536–540.

Boser, B., Guyon, I., & Vapnik, V. (1992). A training algorithm for optimal margin classes. In

Proceedings of the 5th annual workshop on computational learning theory (pp. 144–152).

Springer.

Breiman, L. (2001). Random forest. Machine Learning, 45,5–32.

Chen, X., Cheung, S., So, S., Fan, S., Barry, C., Higgins, J., et al. (2002). Gene expression patterns

in human liver cancers. Molecular Biology of the Cell, 13(6), 1929–1939.

Dagliyan, O., Uney-Yuksektepe, F., Kavakli, I. H., & Turkay, M. (2011). Optimization based tumor

classiﬁcation from microarray gene expression data. PLoS One, 6(2), e14579.

de Souto, M. C. P., Costa, I. G., de Araujo, D. S. A., Ludermir, T. B., & Schliep, A. (2008).

Clustering cancer gene expression data: A comparative study. BMC Bioinformatics, 9,1–14.

https://doi.org/10.1186/1471-2105-9-497

Efron, B., Hastie, T., Johnstone, I. M., & Tibshirani, R. (2004). Least angle regression. The Annals

of Statistics, 32, 407–499.

Fonti, V., & Belitser, E. (2017). Feature selection using Lasso. VU Amsterdam Research Paper in

Business Analytics.

Guo, X., Minai, A. A., & Lu, L. J. (2015). Stable feature selection for clinical prediction: Exploiting

ICD tree structure using Tree-Lasso. Journal of Biomedical Informatics, 27, 277–290.

Guotai, C., Abedin, M. Z., & Moula, F. E. (2017). Modeling credit approval data with neural

networks: An experimental investigation and optimization. Journal of Business Economics and

Management, 18(2), 224–240.

Harb, H. M., & Desuky, A. S. (2014). Feature selection on classiﬁcation of medical datasets based

on particle swarm optimization. International Journal of Computer Applications, 104(5),

14–17.

Holte, R. C. (1993). Very simple classiﬁcation rules perform well on most commonly used datasets.

Springer.

Jovi, A., Brki, K., & Bogunovi, N. (2015). A review of feature selection methods with applications.

In 2015 38th international convention on information and communication technology, elec-

tronics and microelectronics (MIPRO) (pp. 25–29).

Knowledge Mining from Health Data: Application of Feature Selection Approaches 231

Kang, C., Huo, Y., Xin, L., Tian, B., & Yu, B. (2019). Feature selection and tumor classiﬁcation for

microarray data using relaxed Lasso and generalized multi-class support vector machine.

Journal of Theoretical Biology, 463,77–91. https://doi.org/10.1016/j.jtbi.2018.12.010

Kaushik, S., Choudhury, A., & Jatav, A. (2019). Comparative analysis of features selection

techniques for classiﬁcation in healthcare. MLDM, 488–472.

Kira, K., & Rendell, L. A. (1992). The feature selection problem: Traditional methods and a new

algorithm. In AAAI-92 Proceedings (pp. 129–134).

Koh, H. C., & Tan, G. (2005). Data mining applications in healthcare. Journal of Healthcare

Information Management, 19(2), 64–72.

Lu, Y., Yang, L., Shi, B., Li, J., & Abedin, M. Z. (2022). A novel framework of credit risk feature

selection for SMEs during industry 4.0. Annals of Operations Research.

Meinshausen, N. (2006). Quantile regression forests. Journal of Machine Learning Research, 7,

983–999.

Pomeroy, S. L., Tamayo, P., Gaasenbeek, M., Sturla, L. M., Angelo, M., McLaughlin, M. E., Allen,

J. C., Zagzag, D., Olson, J. M., Curran, T., Wetmore, C., Biegel, J. A., Poggio, T., Mukherjee,

S., Rifkin, R., Califano, A., Stolovizky, G., & Louis. (2002). Prediction of central nervous

system embryonal tumour outcome based on gene expression. Nature, 415(436), 436.

Rahman, S., Shahriare, S., Khan, M. I., & Abedin, M. Z. (2021). Risk prediction with machine

learning in cesarean section: Optimizing healthcare operational decisions. In Signal processing

techniques for computational health informatics, intelligent systems reference library (Vol.

192). Springer.

Remeseiro, B., & Bolon-canedo, V. (2019). A review of feature selection methods in medical

applications. Computers in Biology and Medicine, 112, 103375.

Robnik-Sikonja, M., & Kononenko, I. (2003). Theoretical and empirical analysis of ReliefF and

RReliefF. Machine Learning, 53,23–69.

Shipp, M., Ross, K., Tamayo, P., Weng, A., Kutok, J., Aguiar, R., et al. (2002). Diffuse large B-cell

lymphoma outcome prediction by gene-expression proﬁling and supervised machine learning.

Nature Medicine, 8,68–74.

Tibshirani, R. (1996). Regression shrinkage and selection via the lasso. Journal of the Royal

Statistical Society, 58(1), 267–288.

Fintech: Digital Transformation in Finance

Article

Full-text available

Nov 2023

Le Wang

This paper provides an overview of the rapidly expanding fintech sector and its impact on traditional financial services. Fintech has emerged as a customer-centric business model in response to the shortcomings of conventional financial services during the financial crisis and COVID-19. The fintech industry has created prospects in a variety of areas, including blockchain (specifically cryptocurrencies), robo-advisors, online payment services, and peer-to-peer lending platforms. However, the quick expansion of fintech also comes with risks, and investment decisions must be made carefully, considering project duration and volatility. Risk management is crucial to preserving financial stability, and legal oversight is necessary. Machine learning and deep learning techniques can be employed to recognize and mitigate these risks. Banking institutions must take precautions to protect customer data and manage unforeseen operations. The fintech sector has the potential to transform the financial industry while maintaining long-term financial health through a balance of innovation, regulatory monitoring, and financial stability.

Financial Technology and Risk Management

Article

Nov 2023

Yibo Zhen

This essay gives a general overview of how fintech has impacted the financial services sector, highlighting the significance of technologies like blockchain, Robo-Advisors, online payment, and P2P lending. The essay examines how blockchain technology affects financial transactions, including its potential to replace conventional middlemen and improve security and transparency. The rise of Robo-counselors as a disruptive technology that gives investors a cheap and practical substitute for traditional financial counselors is also covered. The article emphasizes the significance of striking a balance between technology and human competence by contrasting the benefits and drawbacks of robot advisors with those of actual advisors. Furthermore, the essay covers the expansion of online payment options, such as digital wallets and mobile payment applications, and how these may change how customers make purchases. Finally, the article explores the potential and hazards raised by the emergence of peer-to-peer lending platforms, which give borrowers an alternative to conventional bank loans. Overall, the article underlines how fintech has the potential to revolutionize the financial services sector, but it also stresses how crucial it is to strike a balance between innovation, financial stability, and regulatory compliance.

DIGITAL TRANSFORMATION AND INTEGRATED FINANCIAL PLANNING: THE FUTURE OF FINANCIAL PLANNING AND ANALYTICS WITH ERP TOOLS

Article

Full-text available

Apr 2024

Marco Fabio Zuccherofino

The paper discusses the impact of digital transformation on businesses and analyzes ERP tools along with the latest technologies. The final chapter focuses on Digital Finance, specifically Digital Financial Planning.

Fraud Detection in Mobile Payment Systems using an XGBoost-based Framework

Article

Full-text available

Oct 2022
INFORM SYST FRONT

Mobile payment systems are becoming more popular due to the increase in the number of smartphones, which, in turn, attracts the interest of fraudsters. Extant research has therefore developed various fraud detection methods using supervised machine learning. However, sufficient labeled data are rarely available and their detection performance is negatively affected by the extreme class imbalance in financial fraud data. The purpose of this study is to propose an XGBoost-based fraud detection framework while considering the financial consequences of fraud detection systems. The framework was empirically validated on a large dataset of more than 6 million mobile transactions. To demonstrate the effectiveness of the proposed framework, we conducted a comparative evaluation of existing machine learning methods designed for modeling imbalanced data and outlier detection. The results suggest that in terms of standard classification measures, the proposed semi-supervised ensemble model integrating multiple unsupervised outlier detection algorithms and an XGBoost classifier achieves the best results, while the highest cost savings can be achieved by combining random under-sampling and XGBoost methods. This study has therefore financial implications for organizations to make appropriate decisions regarding the implementation of effective fraud detection systems.

Auto Checker of Higher Mathematics – an element of mobile cloud education

Article

Full-text available

Mar 2020

We analyzed the main cloud services in the article. We also described the main contribution of mobile cloud technology to education. The article presents the author’s development from the field of mobile cloud education in higher mathematics. The design architecture of this application is described in detail: QR generator and scanner, authorization, sending tasks. Block diagrams and images are presented that clearly demonstrate the operation of the application. We showed an example of solving the integral from the section of integral calculus for higher mathematics and showed how to download the answer in the form of a QR code and find out whether it is correct or incorrect (this can be seen by the color on the smart phone screen). It is shown how this technology helps the teacher save time for checking assignments completed by students. This confirms its effectiveness. Such an application provides students and teachers with the ability to store and process data on a cloud computing platform.

A novel framework of credit risk feature selection for SMEs during industry 4.0

Article

Full-text available

Jul 2022
ANN OPER RES

With the development of industry 4.0, the credit data of SMEs are characterized by a large volume, high speed, diversity and low-value density. How to select the key features that affect the credit risk from the high-dimensional data has become the critical point to accurately measure the credit risk of SMEs and alleviate their financing constraints. In doing so, this paper proposes a credit risk feature selection approach that integrates the binary opposite whale optimization algorithm (BOWOA) and the Kolmogorov–Smirnov (KS) statistic. Furthermore, we use seven machine learning classifiers and three discriminant methods to verify the robustness of the proposed model by using three actual bank data from SMEs. The empirical results show that although no one artificial intelligence credit evaluation method is universal for different SMEs’ credit data, the performance of the BOWOA-KS model proposed in this paper is better than other methods if the number of indicators in the optimal subset of indicators and the prediction performance of the classifier are considered simultaneously. By providing a high-dimensional data feature selection method and improving the predictive performance of credit risk, it could help SMEs focus on the factors that will allow them to improve their creditworthiness and more easily access loans from financial institutions. Moreover, it will also help government agencies and policymakers develop policies to help SMEs reduce their credit risks.

Deep-learning model using hybrid adaptive trend estimated series for modelling and forecasting sales

Article

Full-text available

Jul 2022
ANN OPER RES

Existing sales forecasting models are not comprehensive and flexible enough to consider dynamic changes and nonlinearities in sales time-series at the store and product levels. To capture different big data characteristics in sales forecasting data, such as seasonal and trend variations, this study develops a hybrid model combining adaptive trend estimated series (ATES) with a deep neural network model. ATES is first used to model seasonal effects and incorporate holiday, weekend, and marketing effects on sales. The deep neural network model is then proposed to model residuals by capturing complex high-level spatiotemporal features from the data. The proposed hybrid model is equipped with a feature-extraction component that automatically detects the patterns and trends in time-series, which makes the forecasting model robust against noise and time-series length. To validate the proposed hybrid model, a large volume of sales data is processed with a three-dimensional data model to effectively support business decisions at the product-specific store level. To demonstrate the effectiveness of the proposed model, a comparative analysis is performed with several state-of-the-art sales forecasting methods. Here, we show that the proposed hybrid model outperforms existing models for forecasting horizons ranging from one to 12 months.

'Just Say No' (Or at Least Ask Why) STOMP Medication Reviews in Tower Hamlets Community Learning Disability Service

Article

Full-text available

Jun 2022

Aims 'STOMP stands for stopping over medication of people with a learning disability, autism or both with psychotropic medicines. It is a national project involving many different organizations which are helping to stop the over use of these medicines. STOMP is about helping people to stay well and have a good quality of life'. Our aim was to reduce the percentage of psychotropic burden on the LD and/or autism caseload in Tower Hamlets. Methods We reviewed the internal LD caseload that fit STOMP eligibility criteria (prescribed antipsychotics without an indicated mental health diagnosis). We calculated the% of BNF maximum dose for individual service users, aimed to reduced this, and reviewing the cumulative dose reduction achieved across the service, before and after an intervention. The primary intervention was the introduction of a pharmacy led clinic for service users meeting the criteria. This allowed closer f/u from LD pharmacist, thorough medication histories independent of their routine psychiatric reviews, and using GASS and BAI scales to quantify change achieved to their quality of life. We used early and rigourous people participation to consider the role medications (and their overprescription) in service users quality of life, and asked what service users want out of these medication reviews. Several focus groups were ran without People Participation Lead. Results Prior to starting of clinic - Of 29 STOMP eligible patients within TH CLDS, we have reduced antipsychotics in 8 of them through general raising awareness of STOMP (presentations to staff, reviews of GP letters to identify service users within the caseload who are likely to benefit and/or be receptive to dose reductions etc). So far total reduction of 45.4%, (and a total of three patients have been stopped all together). Conclusion The majority of the results and intervention are yet to be collated, and we are collecting these over the next 2 months, but provisionally we hope to conclude that by reducing the quantity of psychotropic medication we prescribe will improve the quality of life for our service users

The Effect of Flipped Instruction on Special Education Preservice Teachers' Perceptions

Article

Full-text available

May 2022

This study analyzes the flipped instruction model used in three special education educator preparation courses to examine which components preservice teachers perceived most contributed to their content knowledge, motivation, and engagement (n=50). Weekly pre-class asynchronous assignments included the use of educational technology tools such as an interactive e-textbook site, Perusall, and online academic activities such as Khan Academy to strengthen their content knowledge. This allowed more time for a student-centered approach during synchronous instruction to incorporate tools such as Nearpod, Pear Deck, Flipgrid and digital badges to strength-en their motivation and engagement. Data were collected through a post-course survey; results indicate that preservice teachers perceived this model was motivating, engaging, and contributed significantly to their content knowledge. They also identified hands-on activities during class as a significant component of their learning. This article discusses the project, limitations, and implications for future flipped instruction research in special education educator preparation programs.

CHALLENGES OF STUDENT ASSESSMENT IN ONLINE LEARNING

Article

Jan 2023

Overcoming spatial stratification of fintech inclusion: Inferences from across Chinese provinces to guide policy makers

Article

Oct 2022
Int Rev Financ Anal

Few studies have quantitatively analyzed the causes of the uneven development of fintech inclusion. We explore the factors influencing regional differences in China's fintech inclusion by focusing on the influencing factors' spatiotemporal heterogeneity. Fuzzy best-worst, standard deviation ellipse, and geographically temporally weighted regression methodologies were used to investigate the spatiotemporal evolution and influencing factors for 2011–2019. China's fintech inclusion has a higher persistent spatial imbalance in the east and a lower in the central and western regions. Regarding the spatial heterogeneity of influencing factors, we find that the effects of economic development, financial supervision, transportation convenience, population structure and education level show obvious east-west differences. Regarding temporal heterogeneity, we find over our period that the impacts of economic development and financial supervision on fintech inclusion changed from negative to positive, with the positive effects of education level and population structure decreasing. We offer a framework for measuring regional fintech inclusion and practical policy suggestions for promoting coordinated development.

Vocabulary Taught Via Mobile Application Gamification: Receptive, Productive and Long-Term Usability of Words Taught Using Quizlet and Quizlet Live

Conference Paper

Feb 2022

John Senior

Quantifying information transfer among clean energy, carbon, oil, and precious metals: A novel Transfer Entropy-based approach

Article

Jul 2022
Finance Res Lett

Measuring the strength and direction of information flow between markets plays a vital role for investors and policymakers. In this study, we propose a novel approach: the empirical mode decomposition variable lag relative transfer entropy (EMD-VL relative transfer entropy) to assess the pairwise relative causal information transfer over time and scale. Using clean and dirty energy, carbon, and precious metals, we show strong evidence of bidirectional causal information transmission between markets over time-frequency domain.

FinTech Risk Management and Monitoring

Abstract and Figures

Recommended publications

Exploring the Role of Mobile Technologies in Higher Education: The Impact of Online Teaching on Trad...

Model Development for Predicting the Crude Oil Price: Comparative Evaluation of Ensemble and Machine...

An Ensemble LGBM (Light Gradient Boosting Machine) Approach for Crude Oil Price Prediction

The Impact of Big Data on Banking Operations