THE CATHOLIC UNIVERSITY OF EASTERN AFRICA

FACULTY OF SCIENCE

DEPARTMENT OF COMPUTER SCIENCE AND INFORMATION SCIENCE

ONLINE BANKING FRAUD DETECTION SYSTEM

ABIGAEL JEPNG’ETICH KIBET

1049517

A Research Proposal Submitted in Partial Fulfillment of the Requirements for the Award of

Bachelor of Science Degree in Computer Science.

October,2024
DECLARATION AND APPROVAL

I, Abigael Jepng’etich Kibet, declare that this proposal “Online Banking Fraud Detection
System” is my original work and that it has not been presented in any other university or
institution for academic credit.

Signature…………………………………... Date…………………………………………

Abigael Jepng’etich Kibet

1049517

UNIVERSITY SUPERVISOR’S APPROVAL

This research proposal has been submitted for examination with my approval as University
supervisor.

Signature ….………………………… Date …….………………………………

Dr Nicodemus Ishmael

Faculty of Computer Science

HEAD OF DEPARTMENT’S APPROVAL

This research proposal has been submitted for examination with my approval as Head of
Department

Signature ….………………………… Date …….………………………………

Mr Michael Kinyua

Department of Computer Science and Information Science

2
3
ACKNOWLEGMENT
I extend my sincere appreciation to all the researchers before me who conducted valuable
research on which my research will be built on. To my parents for giving me enough space to
work on this document in peace. To that one scammer who almost had me borrowing hustler
fund through phishing and lying that he is from Safaricom, while not using the main Safaricom
call line.

To my two best friends, for casually listening to my rants and mental breakdowns, knowing they
can’t help.

And most of all, to the almighty for guiding and keeping me sane while working on this paper,
knowing that I’m cooked.

4
DEDICATION
I dedicate this research to every single person who has been defrauded by scammers through any
fraudulent activities through the years.

5
ABSTRACT
In the recent years, ranging from 2019 to October 2023, according to the Lexis Risk Solution
Report there has been a 6% rise in financial frauds prior to 12 months, where scams hold at least
35% of the fraud losses. This research proposal presents the development of an online banking
fraud detection system targeting small-scale fraudulent activities, such as phishing, smishing, and
fake online stores. The primary focus is on creating a more interactive and user-friendly
interface, complemented by a robust model for detecting minor frauds in investment firms and an
intuitive dashboard to enhance client understanding. It employs a comprehensive methodology
that consist of questionnaires and surveys for data collection, alongside a review of relevant
articles, journals, and prior research papers to inform the literature and dataset development for
training machine learning models. It utilizes web development technologies like HTML, CSS,
JavaScript, Figma (design tool), and Chart.js for Frontend development, while the backend is
built on Python, Google Colab, MySQL, Scikit-learn, and GraphQL. The key objectives of this
paper include simplifying the user interface to improve user experience, developing machine
learning models suitable for small-scale frauds in investment firms, enhancing customer trust in
investment firms and providing an independent platform for fraud detection and prevention
education. The key outcome is the creation of a UI scanner that evaluates URLs, phone numbers,
emails and SMS messages, that transmits relevant data to the backend for risk assessment and
detailed explanations of fraudulent features. The conclusion underscores that traditional fraud
detection systems often rely on complex data tables and very sophisticated machine learning and
deep learning models, highlighting the necessity for a more specific, accessible and
understandable approach that this system aims to achieve.

6
TABLE OF CONTENTS
Contents
DECLARATION AND APPROVAL.......................................................................................................ii
ACKNOWLEGMENT.............................................................................................................................iii
DEDICATION..........................................................................................................................................iv
ABSTRACT.................................................................................................................................................v
TABLE OF CONTENTS.........................................................................................................................vi
LIST OF FIGURES.................................................................................................................................vii
LIST OF TABLES..................................................................................................................................viii
LIST OF APPENDICESDEFINITION OF KEY TERMS....................................................................ix
CHAPTER ONE......................................................................................................................................11
INTRODUCTION...................................................................................................................................11
1.1 Introduction...................................................................................................................................11
1.2 Motivation and Background of the Research..............................................................................12
1.3 Background of Research...............................................................................................................13
1.4 Problem Statement........................................................................................................................15
1.5 Aim of the Research.......................................................................................................................16
1.6 Objectives of the Research................................................................................................................16
1.6.1 Main Objective of the Research.................................................................................................16
1.6.2 Specific Objectives of the Research...........................................................................................16
1.7 Justification of Research...................................................................................................................17
1.8 Scope of Research..............................................................................................................................18
CHAPTER 2............................................................................................................................................20
LITERATURE REVIEW...................................................................................................................20
2.1 INTRODUCTION.....................................................................................................................20
2.2 REVIEW OF RELATED RESEARCH.......................................................................................20
2.3 REVIEW OF RELATED PROTOTYPES..................................................................................23
2.4 CHAPTER SUMMARY................................................................................................................24
CHAPTER THREE.................................................................................................................................25
RESEARCH METHODOLOGY.......................................................................................................25
3.1 INTRODUCTION.....................................................................................................................25
3.2 RESEARCH DESIGN...................................................................................................................25

7
 3.3 POPULATION..............................................................................................................................25
3.4 SAMPLING TECHNIQUES........................................................................................................26
3.5 SAMPLING INSTRUMENTATIONS.........................................................................................26
3.6 DATA COLLECTION..................................................................................................................27
3.7 DATA ANALYSIS TECHNIQUES..............................................................................................28
3.8 LIMITATIONS..............................................................................................................................29

LIST OF FIGURES

8
LIST OF TABLES

Table 1: Sample Instrumentations & its purposes.....................................................................................25

Table 2: Data collection methods & its purposes......................................................................................26

9
LIST OF APPENDICES

10
DEFINITION OF KEY TERMS

Online Banking: It is defined as a convenient system that allows you to conduct various
financial transactions through the internet.

Fraud: It’s the wrongful or criminal deception intended to result into financial or personal gain.

Fraud Detection System: They are analytical tools used to identify and prevent fraudulent
activities by analyzing patterns in data and flagging anomalies using various algorithms and
models to assess transactions or behaviors, enabling the making of informed decisions using
data-driven insights.

User Interface: The visual elements and interactions that allow a user to interact with a
computer program or device.

Phishing: A type of cybercrime attack, where attackers pose as a trustworthy entity in an attempt
to trick individuals into revealing personal information like passwords and credit card numbers.
Can be done through text as smishing and calls as vishing.

Machine Learning: It’s a subset of Artificial Intelligence where machines are taught to learn
from supervised and unsupervised data.

Risk Assessment: A systematic used to identify and analyze potential hazards and risks in a
situation.

Interactive Dashboards: They are powerful tools for visualizing and analyzing large datasets
and providing real-time insights into suspicious activities during fraud detection.

Interactive Data Tables: They are visual representations of structured data that allows users to
interact with the information directly.

Small-scale Fraudulent Activities: It refers to deceptive actions or schemes carried out on a

relatively limited or localized basis. It targets small businesses, individuals and specific
transactions. e.g. Ponzi schemes and phishing.

Investment Firms: It’s a specialized business entity engaged in managing, selling, and
marketing investment products to the public. It handles pooled capital from investors and invests
it in various financial securities.

11
 CHAPTER ONE
INTRODUCTION
1.1 Introduction
The Online Banking Fraud Detection System represents a pivotal advancement in Small-scale
Fraudulent activities, particularly in addressing the critical fraud issues in Investment Firms. In
the recent years, ranging from 2019 to October 2023, according to the Lexis Risk Solution
Report there has been a 6% rise in financial frauds prior to 12 months, where scams hold at least
35% of the fraud losses. This rise has not only caused a major landslide in financial losses but
also led to a negative impact on customer trust. And as we all know maintaining customer trust is
crucial for investment firms. And with rise of sophisticated and impersonal fraudulent attacks
due to the digitization and automation of financial systems, there has been a need for an
innovation to ensure the security of investments and reduction of financial losses.

The core problem underlying this fraud detection gap is the use of complex data table and highly
sophisticated machine learning models used to build the fraud detection systems that are
currently in use. These systems are encompassed with challenges which include; risk of
information overload and intensiveness caused by use of interactive data tables for UI design,
limited contextual insights and the generalization of fraud detection systems, therefore making
some of the fraudulent activities go unnoticed due to the increased sophistication of fraudsters.
And with the increasing digitization of money transactions, there has been an increase in
vulnerability of user data security that can be easily acquired through phishing. This can
therefore lead to identity theft once sensitive private personal details have been acquired. Hence,
financial losses are then registered.

The proposed Online Banking Fraud Detection System aims to offer an alternative solution to
investment firms on how to handle small scale fraudulent activities like phishing while
maintaining customer trust. By developing an independent platform that acts a proofreading tool
for fraud, investment firms can not only build trust but also attract more clients who are willing
to place their funds under these firms. This innovative model is expected to enhance trust,
efficiency, security and reduce financial losses.

12
The development and implementation of the online banking fraud detection section system the
data security guidelines which enable the security of finances. By providing a fraud detection
and fraud prevention platform, investment firms are bound to retain customer trust and reduce
financial losses globally.

In conclusion, the introduction of the online banking fraud detection system for small scale
fraudulent activities, will enable investment firm to be more prepared for sophisticated
fraudulent activities against them. As they can double check for fraud using their main system
and proofread using this system. This will also keep their relations with their clients intact as the
customer will feel more secure knowing that there’s a two-step verification for fraud before
certain investments are made.

1.2 Background of Research

With a notable rise of 11% of reported fraud crimes that occurred in investment firms between
the years 2018 to 2024, it's logical to say that the current fraud detection systems aren't really
built specifically for certain fraudulent practices and are more complex and generic. Where
Investment firms notably manage a substantial number of assets, a reliable fraud detection
system is required.

The body of literature on fraud detection in online banking has evolved significantly from 2019
to 2024. Recent studies emphasize the use of machine learning techniques to enhance fraud
detection capabilities. For instance, Akhter et al. (2021) demonstrated that supervised learning
models, particularly gradient boosting and random forests, significantly improved the accuracy
of identifying fraudulent transactions compared to traditional methods.
The research has also shown the unique characteristics of various fraud types. For instance,
Ponzi schemes, which often rely on a lie of legitimacy, require sophisticated detection methods
that analyze network structures and behavioral anomalies (Sajid et al., 2022). In terms of
phishing, studies have shown that using both machine learning with natural language processing
can effectively identify malicious emails and URLs, thus reducing successful phishing attempts
(Singh & Gupta, 2023). Additionally, identity theft detection has seen advances through the
application of anomaly detection techniques, which focus on identifying changes in user
behavior (Hussain et al.,2020).

13
Despite these advancements, there remains a notable gap in integrated systems capable of
simultaneously addressing these various types of small-scale fraud. Many systems still work in
generalization and complexities, limiting their effectiveness in fully detecting fraud in dynamic
online environments. The following theoretical models will be studied when developing this
online fraud detection system for small scale fraudulent practices in investment firms:

Fraud Triangle Theory: It's a theoretical framework that underlines that fraud occurs when
three elements; opportunity, motivation, and rationalization occur. By understanding these
elements, the research can be informed on the design of detection algorithms that target high-risk
transactions.
Behavioral Economics: This framework can be used to understand user decision-making
processes and how psychological factors may contribute to the chances of fraud occurring,
particularly in scenarios like phishing and identity theft.

Anomaly Detection Theory: Is a theory that is vital for identifying unusual patterns in
transactional data, which is particularly relevant for detecting identity theft and phishing
attempts. The use of machine learning algorithms for anomaly detection will be central to the
proposed system.
Risk Scoring Models: They are models that assign risk scores to transactions based on multiple
factors, such as transaction amount, frequency, and user history, to identify potentially fraudulent
activities.
The proposed online banking fraud detection system will have significant effects on investment
firms and their clients. By providing targeted detection capabilities for small-scale fraud, the
system will help protect client assets, reduce financial losses, and enhance trust in online banking
platforms. Furthermore, this research proposal will contribute to the broader academic discourse
on fraud detection methodologies and cybersecurity practices within the financial sector,
specifically in investment protection. By addressing the common vulnerabilities in small-scale
fraudulent activities, this work aims to create a more secure financial ecosystem, ultimately
benefiting all stakeholders involved.

14
1.3 Problem Statement
The financial sector like investment firms have embraced digital technologies such as online
banking to enhance convenience, cost saving, efficiency and security for individuals and
businesses alike. What wasn't expected was the insecurity that online banking brings. And with
the acceptance and increasing use of online banking, small scale fraudulent activities such as
Ponzi schemes, phishing and identity theft have been encountered. Investment firms which
manage a substantial number of assets have been a major target for sophisticated fraudsters in the
financial sector which has led to great financial losses and a loss in customer trust. Since 2018
there has been a notable increase in fraud affecting the investment sector, where a definitive rise
of 11% in fraudulent crime reports have been registered. Current fraud detection systems often
focus on large-scale frauds or employ complex and generic algorithms that may not be well
tailored to identify subtle small-scale fraudulent activities. This research proposal aims to
develop a specialized online banking fraud detection system that effectively identifies and
mitigates these specific small scale fraudulent practices within investment firms.

1.4 Aim of the Research

This research proposal aims to provide, develop, and implement an independent online banking
fraud detection system that addresses a critical gap in investment firms concerning small scale
fraudulent practices. The system will act as a two-step verification and proofreading tool for
fraudulent activities that may go unnoticed at first, thus cementing customer trust in investment
firms and reducing financial losses by a mile. The proposed system seeks to offer a far simplified
user interface for better understanding, a more robust machine learning model that has been fine-
tuned to detect small scale fraudulent practices, specifically Ponzi schemes, phishing and identity
theft. By developing this system, there will be enhanced efficiency, effectiveness, accuracy and
security of assets being held by investment firms.

15
1.5 Justification of Research
Based on data collected by The Federal Trade Commission (FTC), consumers have reported
losing nearly $8.8 billion to various online scams in 2022, with phishing attacks, identity theft
and Ponzi schemes being significant contributors.

Investment scams, like Ponzi schemes account for loses above $3.8 billion, marking a substantial
increase from the previous years. The FTC’s data highlights the growing threat of these types of
frauds and the need for increased vigilance and preventative measures. And with the
sophistication of technologies, scammers have equally become sophisticated to keep up with the
new technologies. Where some use deep fake AI to scam people into investing in them by faking
to be a high profile business mogul or an A-list Celebrity.

Focusing on Kenya, the Central Bank of Kenya (CBK), in 2021 registered that 6.1% of digital
banking platform users, 29.5% of mobile money users like M-pesa and 6.8% of bank account
uses reported a loss of money caused by fraud and the lack of its detection. The PwC Global
Economic Crime and Fraud Survey 2022 showed that fraudsters in Kenya have adapted to the
technological trends. This adaption has therefore led to a rise in digital financial problems.

With the sophistication of online fraudsters, the current fraud detection systems can keep up.
Being that most of the current fraud detection systems for online banking are rule based systems,
where these systems have predefined rules to identify potentially fraudulent activities. With a
new introduction of fraudulent methods like the use of deep fake AI, some frauds go unnoticed.
Similarly, the current systems are highly generalized instead of being built to focus on specific
fraudulent activities, to reduce undetected fraud.

These discoveries have therefore led to the development of the proposed online banking fraud
detection system, where machine learning will be employed in the building of the system
because of its learning nature. The machine learning model of the proposed system will not only
detect fraudulent activities; it will also learn from the data tested on it to improve its detecting
abilities overtime. The proposed system also aims to focus on small-scale fraudulent activities
specifically rather than focusing on a general perspective in investment firms. However,
although machine learning has been used to predict credit card and financial statement frauds,
loan defaults, and money-laundering transactions, no researchers to date have employed machine

16
learning to predict fraud in financial markets (Duman and Ozcelik, 2011, Perols, 2011, Perols et
al., 2017, Sahin and Duman, 2011). By introducing a novel machine learning technique into the
process of financial market regulation, I contribute to the scholarship on critical fraud research.

1.6 Objectives of the Research

1.6.1 Main Objective of the Research
To develop an Online Banking Fraud detection system

1.6.2 Specific Objectives of the Research

i. To simplify user interface and improve user experience by making use of interactive
dashboards for results.
ii. To develop a more robust machine learning model that is fine tuned into catering for
small scale fraudulent activities such as Ponzi schemes, phishing and Identity theft.
iii. To provide an independent platform that acts a two-step verification and proofreading
tool for unnoticed fraudulent anomalies thus reducing financial losses.
iv. To enhance customer confidence and trust in investment firms.

1.7 Motivation and Background of the Research

Before choosing an investment to make, most investors want an investment where they will
make high returns. Most fraudsters know about this and they take advantage of it. This is where
Ponzi schemes come in. A Ponzi scheme is like a magician’s trick gone awry. It’s all about
luring unsuspecting investors with promises of dazzling results, but behind the curtain it’s all
smoke and mirrors. Basically there’s high returns and low risk in Ponzi schemes making it very
enticing for a new investor who doesn’t know much about such fraudulent activities.

Like how phishing is to an individual, where I’m coerced into giving personal sensitive
information, Ponzi schemes are a corresponding problem to investment firms. A good example
of a Ponzi scheme that costed people lots of money is:

17
The Madoff Victim fund: It orchestrated a $64 billion security fraud Ponzi scheme. The turn
tables of this scheme is that the MVF distributed over $158.1 million of the total amount to its
25,000 victims worldwide which ranged from wealthy investors to charities and pension funds as
restitution. In December 2023, MVF restituted over 40,800 of its victims and kept the rest of the
amount. It’s considered the most financial crimes in history.

And as we traverse through the years, we find that investment fraud has unprecedented heights.
In 2023, the number of investment fraud victims skyrocketed 11 times higher than five years
prior. Each victim faced an average loss of $115,449, up from $68,496 in 2018. And with most
financial transactions being made online, most customers and financial institutions are at risk for
fraud. Keeping in mind that investment firms handle substantial amount of money and sensitive
personal information, which are very vulnerable to online banking. And with a surge in phishing,
identity theft and Ponzi schemes, financial losses are taken and customer trust is lost.

The motivation of research of the proposed system stems from the urgent need for effective fraud
detection systems built for the unique challenges faced by investment firms. As financial markets
grow more complex and cybercriminals become increasingly sophisticated, traditional methods
of fraud detection are proving to be inadequate. By exploring innovative solutions that leverage
technologies such as machine learning and data analytics, this research aims to enhance the
detection of small-scale fraudulent activities and improve overall security in online banking
environments.

Taking into consideration that investment firms operate on security and trust, the impact of fraud
can be quite devastating. Not only in terms of financial losses but also in reputational damage
that could deter current clients and potential ones. Many firms are making use of fraud detection
systems to safeguard assets and maintain customer confidence. These systems often utilize
various data sources and employ algorithms to analyze transaction patterns, identify anomalies,
and flag potentially fraudulent activities in real time. However, challenges remain in
implementing effective detection systems that balance accuracy with efficiency. And this is
where my proposed system comes in. By developing and fine-tuning a model to specifically
detect anomalies for small scale fraudulent activities in investment firms like phishing and Ponzi
schemes accuracy and efficiency can be achieved.

18
1.8 Scope of Research
The scope of the research on Online Banking Fraud Detection System delves into several
dimensions, this research is dedicated in addressing the critical gaps in fraud detection within
investment firms when it comes to online banking. This is done through the development and
implementation of a Specified online banking fraud detection system for small scale fraudulent
activities. The following consists of the scope of the study:

i. To design, develop and implement an efficient, effective and simple specialized system
that uses robust machine learning models to detect small scale fraudulent activities during
online banking in investment firms.
ii. The research focuses on Investment firms as the target organizations for the deployment
of the system, these include Venture Capital Firms, Real Estate Investment Trust, Unit
Investment Trust and Private Equity firms that can offer investment services.
iii. The proposed system will include several key features:

A user interface scanner, where you can upload URLs, SMS and certain documents about
investments you’re interested in to ensure their eligibility.
A secure database for keeping a record of the scanned data that can be used to constantly
retrain the machine learning model.
A machine learning model that has been developed and fine-tuned to detect a specific
forms of fraudulent activities like Ponzi schemes, phishing and identity theft.
An interactive dashboard that will act as a simple explainable feature to show the
distribution of what could likely be a fraudulent pattern.
iv. The implementation of the system in selected investment firms, the characteristics of the
target population include startup companies, existing businesses and real estate investors
who make use of online banking platforms as a way of tracking their investment.
v. The research of the system will be conducted in firms that will be selected based on high
chance of financial losses, increased complaints of fraudulent activities and adequate
infrastructure for implementing the system.
vi. The research of the system will approximately take 3 to 5 months; the approximated time
period will ensure adequate data is reviewed for a sound research to be recorded.

19
vii. The study will be built on theories related to small scale fraudulent activities, machine
learning algorithms for fraud detection and online banking in investment firms. It will
explore the impact of seamless independent proofreading system for Ponzi schemes,
phishing and identity theft; thus reducing financial losses.

1.9 RESEARCH QUESTIONS

1. Do the fraud detection systems built on machine learning solutions offer effective
detection of small-scale fraudulent activities as compared to traditional rule-based
systems?
2. Does the behavioral analysis of user patterns impact the detection of online banking
frauds?
3. Do complicated data tables make it difficult for cybersecurity analysts and financial
analysts to detect fraudulent activities in their institutions?
4. Does the lack of a proofreading fraud detection system impact the rise of undetected
fraudulent activities globally?

1.10 ALTERNATIVE HYPOTHESES

1. A simplified user interface in online fraud detection systems will significantly improve
the detection rate of phishing attempts and reduce response time within financial
institutions compared to traditional, complex user interfaces.
2. Implementing machine learning in online fraud detection systems will reduce the false
positive rate and decrease time rate for detection in identity theft detection by at least
15% compared to existing systems used by financial institutions like rule-based systems.
3. Designing the fraud detection system as a "proofreader" — one that continuously
monitors and cross-checks financial transactions and activities for signs of Ponzi schemes
— will reduce the occurrence of undetected fraud by ensuring greater vigilance and
accuracy in the detection process.

20
 CHAPTER 2
LITERATURE REVIEW
2.1 INTRODUCTION
Online banking fraud detection system are very crucial in safeguarding digital transaction which
have become the main target for fraudsters worldwide. It helps in maintaining e-commerce trust
all around the world as businesses have opted to digitize their functionalities. With rapid growth
of online transactions caused by these sudden changes with businesses, the risk of fraud has
increased significantly, rising the need for advanced detection mechanisms.

Early fraud detection systems relied on rule-based algorithms and manual reviews. However, the
complexities and increased volume of financial online transactions have driven the adoption of
machine learning and artificial intelligence techniques for detection. These advanced methods
can analyze large amounts of data, identify patterns and detect anomalies more effectively than
traditional methods.

As stated in the first chapter, my system aims to build an independent platform that acts as a
proofread system for fraud detection. It aims to catch patterns that may have been missed. It is
meant to be built om machine learning and AI techniques alongside a simplified User Interface.
In this chapter, I will support and show why my system is required and how it correlates to all
the previous systems built.

2.2 REVIEW OF RELATED RESEARCH

Based on a financial report conducted in East – Asia in Australia, Hong Kong and Singapore, it
shows that whilst all countries all around the world are going through an increasing level of
scams and frauds, richer countries in the west like G7, EU & OECD member countries are
reporting the most in terms of cases & losses. A few countries or jurisdictions are reporting
sufficient information to be able to understand the threat well enough & appreciate the response.
These include Australia, Hong Kong SAR & Singapore, which this report is focused on, as well
a handful of others, for example New Zealand, Taiwan, Canada, UK & USA.

Proceeds from reported losses from Australia, Hong Kong SAR and Singapore (as well as
reported losses from other countries with reported losses, namely the UK, Canada & the USA) if
extrapolated globally, generate estimates between US$50 Billion and US$177 Billion with a

21
mid-point estimate of US$114 Billion, which would make it the 64th largest country by GDP
equivalent. Average losses per victim are estimated therefore at US$12,000 and average losses
per citizen at approximately US$62. Despite this, scam losses are widely acknowledged as being
under reported.

Factors which make countries or jurisdictions more attractive targets, include those that are
richer but also, more advanced in terms of digitalization & connectivity, are able to offer digital
remote banking & faster payments, with speedy remote opening of many new bank accounts, use
of none cash including use of virtual currencies, significant online time being spent including on
the internet & on social media & the availability & popularity of e-commerce, having languages
spoken common to those of the scammers & of course a general limited awareness of enough
citizens to scam types & tricks & how to defend themselves.

According to Interpol’s Secretary General Juergen Stock in his assessment on global financial
fraud published in March 2024, “we are facing an epidemic in the growth of financial fraud,
leading to 1 individual, often vulnerable people, & companies being defrauded on a massive and
global scale. Changes in technology & the rapid increase in the scale & volume of organized
crime has driven the creation of a range of new ways to defraud innocent people, business &
even governments. With the development of AI and Cryptocurrencies, the situation is only going
to get worse without urgent action. It is important that there are no safe havens for financial
fraudsters to operate. We must close existing gaps & ensure information sharing between sectors
& across borders is the norm, not the exception. We also need to encourage greater reporting of
financial crime as well as invest in capacity building & training for law enforcement to develop a
more effective & truly global response.”

And based on the information detailed above, the main question arises, “With the rise of
digitization of everything all across the globe and its rapid growth, is it possible that some
fraudulent activities such as identity theft, Ponzi schemes and phishing which are all small-scale
fraudulent activities still go unnoticed due to lack of a highly specific fraud detection system?”
or “Is it due to the lack of proofreading from expert handling such situations?”. And while
majoring in the investment sector, where lots of financial transactions are carried a lot of
vulnerabilities are seen to occur if not highly monitored.

22
The following is a comparative study on online banking challenges from South Africa and Spain.
Undoubtedly, both banks, investment firms and customers benefit in using online banking
services. However, there are also challenges in ensuring security and privacy to fight against
fraudulent activities. The banking sector and financial institutions endure yearly losses through
online banking fraudulent crimes (Kumar, Mubarak & Dhanush 2020). The annual crime
statistics shows that although the online banking fraud makes up the smallest portion of incidents
of digital banking crime (20% of reported incidents), it accounts for the second highest portion of
gross losses at 45% (SABRIC 2021).

Based on reports and yearly statistics that have been conducted, it shows that online banking
fraud crime allegations are rising every year with the reported incidents reaching 20,000,000 in
2021. It is important to note that when compared to bank application and mobile banking fraud,
online banking fraud reflected the highest average financial value per incident. The online
banking services risks and challenges must be mitigated to foster and preserve customer’s trust.
Online banking fraud has negative repercussions for all parties affected, this includes financial
cost, inconvenience and loss of trust. From the literature review, studies identified few
challenges related to online banking security that could result in fraud as discussed below:

Lack of strong authentication – researchers have highlighted the necessity of having additional
security measures to authenticate customer’s identity on online banking application. When the
same passwords are frequently used for many services, fraud vulnerability increases whenever
such information is stolen (Shankar & Rishi 2020)
Humans compromising security – customers and bank employees can both be considered as the
weakest links in fraud. Several peoples use their own gain and may cause loss to both customers
as well as the bank. Occasionally, internal bank employees purposefully compromise security to
commit fraud (Ranjith 2019).
Online banking fraud generally has the common deceptive characteristics of cybercrime
perpetrated in a business relationship for personal gains. It is an act to intentionally deceive
through false information, claim or the suppression of the truth (Crouhy, Galai & Wiener 2021).
The literature reveals that the most common type of online fraud in the banking sector is
phishing. Generally, phishing could be defined as a scalable act of deception whereby
impersonation is used to obtain information from a victim (Arianna et al. 2022).

23
Based on the literature drawn from previous research papers above, it is safe to say, by building
more specific systems to cater towards small scale fraudulent activities we would be able to
reduce the number of fraudulent activities worldwide.
2.3 REVIEW OF RELATED PROTOTYPES
The following are some online banking fraud detection systems and prototypes:

Rule-based systems are a traditional approach to decision making and problem solving, often
used in fraud detection. It uses predefined rules to make decisions or solve problems. The rules
are usually in the form of “if-then” statements that can dictate the nest actin to be taken when
certain conditions are met. It is used in fraud detection to identify suspicious transactions based
on a predefined criterion. It is built on a knowledge base, interference engine and a working
memory.

Behavioral biometrics is a security technology that analyzes a user’s unique behavior patterns to
verify their identity and detect fraudulent activities. It analyzes a user’s digital, physical, and
cognitive behavior to distinguish between legitimate users and cybercriminals. It uses machine
learning to analyze patterns in human activity and detect whether someone is who they claim to
be when interacting online. It is mostly in use in Banking & Financial Services, E-commerce,
Telecommunication and Healthcare.

The Amazon fraud detector is a machine learning fraud detection system offered by Amazon
Web Services (AWS) designed to help detect potentially fraudulent activities online such as
unauthorized transactions and the creation of fake accounts. It has been using machine learning
for fraud detection for over 20 years to help businesses identify fraud. It has been built on
machine learning models that have been trained on vast amounts of historical fraud data. It uses
advanced ML techniques to analyze data and identify patterns that indicate fraudulent activities.

Anomaly detection systems are used to identify unusual patterns that do not conform to expected
behavior within a dataset. It helps in spotting outliers. These anomalies can indicate critical
incidents like fraud. It is mostly used in e-commerce and banking for fraud detection. It is built
on statistical methods, machine learning algorithms, and deep learning techniques.

All the systems above have certain issues that have made it a need to build another system. These
challenges include: Rule-based systems are not scalable and are rigid. Behavioral biometrics are

24
built for complex frauds and may overlook simpler ones. Amazon fraud detector is built for users
on AWS. While, Anomalies detection systems depend on data, so if the data quality isn’t up to
par, there may be a lot of false positives and false negatives.

2.4 CHAPTER SUMMARY

Online banking fraud detection systems play a crucial role in safeguarding digital transactions,
which have become major targets for fraudsters as businesses increasingly digitize. The rapid
growth of online transactions has led to a significant rise in fraud, prompting the need for
advanced detection methods. Early fraud detection relied on rule-based algorithms and manual
reviews, but with the growing complexity and volume of transactions, machine learning (ML)
and artificial intelligence (AI) have become more effective tools. These technologies can analyze
large data sets, identify patterns, and detect anomalies better than traditional systems. A report on
financial fraud in regions like Australia, Hong Kong, and Singapore reveals that fraud losses
globally could range from $50 billion to $177 billion, with average losses per victim at $12,000.
Factors like high digitalization, fast payments, and limited public awareness make wealthier and
more connected countries prime targets for fraud. The rise of AI and cryptocurrencies further
complicates the situation, increasing the need for global cooperation, enhanced information
sharing, and better reporting and law enforcement responses. Despite these advancements,
challenges remain in detecting smaller-scale fraud like identity theft and phishing. Studies in
South Africa and Spain indicate that online banking fraud, although a smaller portion of digital
crimes, results in significant financial losses. Factors such as weak authentication, human error,
and internal fraud contribute to these risks. Common fraud techniques, particularly phishing,
exploit these vulnerabilities. Current fraud detection systems, such as rule-based systems,
behavioral biometrics, Amazon’s fraud detector, and anomaly detection, each have limitations.
Rule-based systems are rigid and non-scalable, while behavioral biometrics may miss smaller
frauds. Amazon's fraud detection is limited to AWS users, and anomaly detection relies on high-
quality data to avoid false positives and negatives. These issues highlight the need for more
detailed systems that can detect a wider range of fraudulent activities, especially smaller ones
that often go unnoticed.

25
 CHAPTER THREE
RESEARCH METHODOLOGY
3.1 INTRODUCTION
This chapter focuses on the detailed description of all the techniques that would be used to delve
into the research of this proposal. As stated in the first chapter, the online banking fraud
detection system focuses on building a system with a front-end and a back-end. It will be built on
machine learning models that will work effectively and efficiently as compare to previous
systems in this field. It will majorly focus on the research design, whether quantitative or
qualitative or both, the population where this research will affect. It will also suggest the best
sampling techniques and instrumentations. It will also delve into the nitty gritty of data collection
methods alongside it’s limitations that will suffice this research for a good research conduction
and preparation.

3.2 RESEARCH DESIGN

The research design for this study on online banking fraud detection systems will be a mixed-
methods approach, integrating both qualitative and quantitative methods. The design chosen
provides a comprehensive analysis of the current state of fraud detection systems, their
effectiveness, and areas of improvement.

The quantitative approach will make use of a survey to gather data from investment firms, banks
and online banking customers about their experiences with fraud detection systems. A qualitative
approach will focus on in-depth interviews and questionnaires to explore perceptions and
insights of investors about their experience with fraudulent activities such as phishing and Ponzi
Schemes.

3.3 POPULATION
The population we aim to focus on for this study includes:

Investment firms: They are the main case study for this system, hence they are our main
population. Here we shall focus on financial analysts who are responsible for analyzing
investments opportunities and assessing financial risks. They are the most likely to encounter
fraud in their settings. Alongside them we will also focus on IT security specialist who manage
and secure the firms digital infrastructure against fraudulent activities.

26
Banks: They work in a close knit with investment firms and online banking customers who are
our main targets. We shall focus on Fraud detection analysts who dedicate their time and skills to
identify and investigate fraudulent transactions. Alongside them we shall also take Customer
service representatives who interact with customers and handle reports of suspected fraudulent
cases.

Online Banking Customers: In this sector of the population, our main focus will be frequent
users, senior citizens and small business owners. Customers who use online banking frequently
are more susceptible to fraud, similarly senior citizens are also liable victims. For senior citizens,
due to their ages they are considered easy to fool and have quite a lot to lose. As for small
business owners, may fall for Ponzi schemes which promise high returns at little to no risk,
which may result into being scammed out of their money.

3.4 SAMPLING TECHNIQUES

The sampling techniques that would be use would be the following:

Stratified Random Sampling: It ensures a balanced representation across different strata like
different investment firms, bank sizes and diverse customer demographic. The population will be
divided into subgroups and samples will be drawn randomly from each subgroup to prevent bias.
This will mainly focus on the online banking customers who are frequent user, senior citizens or
small business owners.

Purposive Sampling: It involves selecting experts and key stakeholders within the investment
firms and banks based on their expertise in their field and relevance to fraud detection. This will
majorly focus on Financial analysts and IT security specialists in investments firms alongside
Fraud detection analysts and Customer Service representatives in banking institutions.

3.5 SAMPLING INSTRUMENTATIONS

The following sampling instrumentation tools will be used:

Table 1: Sample Instrumentations & its purposes

Sampling Purpose Contents

Technique
Surveys To gather quantitative data from Asking Demographic Questions like age,
a large number of respondents. occupation etc.

27
 To identify trends and patterns in Usage patterns of online banking.
online banking fraud experiences Experiences with online banking fraud.
and perceptions. Perception of current fraud detection
To collect feedback on current system.
fraud detection system. Suggestions for improvements.
Questionnaire To obtain detailed, structured Structured questions similar to surveys.
s information from respondents. Multiple choice –like questions.
To quantify attitudes and Specific scenarios and user reactions.
behaviors related to online Will have open-ended questions for a
banking fraud. more detailed feedback.
To understand the effectiveness
of specific fraud detection
methods.
Interviews To gather qualitative data Will have open-ended questions.
through detailed personal Follow –up questions for clarifications.
insights for explaining more on Questions on specific fraud incidents and
the survey context for complex response methods.
issues. In-depth explorations of expert opinions
To get expert opinions and and recommendations on fraud detection
recommendations on bettering systems.
fraud detection systems.

3.6 DATA COLLECTION

The following are the primary and secondary data collection methods and their purposes:

Table 2: Data collection methods & its purposes

DATA TYPES PURPOSE

COLLECTION
METHOD
PRIMARY Interviews To gather qualitative data through detailed personal
insights for explaining more on the survey context for

28
 complex issues.
To get expert opinions and recommendations on
bettering fraud detection systems.
Surveys To gather quantitative data from a large number of
respondents.
To identify trends and patterns in online banking fraud
experiences and perceptions.
To collect feedback on current fraud detection system.
Questionnaires To obtain detailed, structured information from
respondents.
To quantify attitudes and behaviors related to online
banking fraud.
To understand the effectiveness of specific fraud
detection methods.
SECONDARY Literature To gather existing knowledge and theories.
Review To identify gaps in current research.
To provide a theoretical framework for the study.
Financial To understand the economic impact of online banking
Reports fraud.
To identify patterns and trends in fraud incidents.
To provide empirical data to support the research.

3.7 DATA ANALYSIS TECHNIQUES

The following data analysis techniques will be used:

Descriptive Analysis: This will majorly be used for quantitative data that would be acquired from
surveys. Using tools like MS Excel and Power BI, we will be able to summarize and understand
historical data. We can therefore derive data visualizations and summary statistics from surveys,
literature review and financial reports.

29
Predictive Analysis: It will help in predicting the likelihood of future fraudulent activities based
on historical data. Machine learning models such as logistic regression and neural networks will
be used. It will be used to score transactions based on fraud risk and predict future fraud
incidents. This method of analysis will mainly be focused towards datasets or generated datasets.

Thematic Analysis: Will be used to interpret and analyze textual data from questionnaires and
interviews. It’s a valuable method in understanding the underlying patterns, themes and trends
within the qualitative data in this research.

Content Analysis: It is used to quantify and analyze the presence, meanings and relationships of
certain words, themes or concept with qualitative research. It will be used in recognizing
relationships among words and themes in financial reports and literature reviews in this research.
It will be used alongside thematic analysis for an increased accuracy and effectiveness.

3.8 LIMITATIONS
The following are some limitations that may be encountered during the research:

Response Bias: While during research, we send out surveys and questionnaires to be filled. And
sometimes the responses and feedbacks given may lack accuracy which is caused by the
respondent’s willingness to offer truthful responses due to subjective reasons.

Access to quality data: During research, it is essential that the datasets and data being used are of
high quality and are relevant to the research. And for an even higher accuracy of models to build
the system, we must have a large number of high quality datasets, which is difficult as financial
data is very confidential and highly protected. It will require a long process that may take
unnecessary time to be granted access to such data.

Rapid change in technologies: The fast evolution of fraud detection technologies may lead to
findings become outdated quickly.

30
REFERENCES
https://sajim.co.za/index.php/sajim/article/view/1763/2689
https://www.tandfonline.com/doi/full/10.1080/15564886.2023.2263756
https://www.researchgate.net/publication/376827420_Banking_Fraud_Identification_and_Preve
ntion

https://www.sciencedirect.com/science/article/pii/S1319157823000034
https://www.sciencedirect.com/science/article/pii/S0167404823002973

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9940837/

https://www.gcffc.org/wp-content/uploads/2024/06/GCFFC-Scams-Report-6June2024Pbd-3.pdf

https://www.mdpi.com/2076-3417/12/19/9637

https://www.frontiersin.org/journals/computer-science/articles/10.3389/fcomp.2021.563060/full

Paladini, T., Bernasconi de Luca, M., Carminati, M., Polino, M., Trovò, F., & Zanero, S. (2
023). "Advancing Fraud Detection Systems Through Online Learning." In Machine Learning an
d Knowledge Discovery in Databases: Applied Data Science and Demo Track (pp. 275-145. T
his review maps the literature on banking sector frauds, covering various databases and publicati
ons from 2000 to 20214.Online Banking Fraud Detection System: A Review. International J
ournal of Advanced Trends in Computer Science and Engineering, 8(1), 45-58. This review pa
per provides an overview of different fraud detection systems used in online banking5."A Syste
matic Literature Review of E-Banking Frauds: Current Scenario and Security Techniques
" by Iftikhar Ahmad, Shahid Iqbal, Shahzad Jamil, and Muhammad Kamran2. This review explo
res the current scenario of e-banking frauds and discusses various security techniques to combat
them2."Guardians of the vault: unmasking online threats and fortifying e-banking security
, a systematic review" by Iftikhar Ahmad, Salim Khan, and Shahid Iqbal1. This paper investiga
tes the adoption of digital technologies in the banking industry and its impact on the rise of digita
l fraudulent activities, including Ponzi schemes, phishing, and identity theft1.

31