VISVESVARAYA TECHNOLOGICAL UNIVERSITY

JNANA SANGAMA, BELAGAVI - 590018

A Synopsis on
ONLINE PAYMENT FRAUD DETECTION SYSTEM

Submitted for the Final Year Project of the AY: 2024-25

by

Aditya Chug (1NT21EC004)

Ruchi Yadav (1NT21EC117)
Shreyas Somanache (1NT21EC148)

Under the Guidance of

Prof. Prajna.K.B
Associate Professor
Dept. of Electronics and Communication Engineering

DEPARTMENT OF ELECTRONICS AND COMMUNICATION ENGINEERING

YELAHANKA, BENGALURU- 560064
 Contents
List of Figures 1
List of Tables 1
1. Introduction 2
2. Literature Survey 3-4
3. Motivation and Objectives 4-5
3.1 Motivation 4
3.2 Objectives 5
4. Architecture of the Proposed System 6-7
5. Algorithm Description 8-9
6. Design Specification/Dataset Description 9
7. Expected Outcomes 10-11
References 12
Remarks 13

i
1. Introduction
Detecting online payment fraud is a critical endeavour that demands a comprehensive
approach encompassing various stages, from data collection to continuous monitoring and
improvement. The initial stage entails the meticulous compilation of a comprehensive
dataset, encompassing both legitimate and fraudulent transactions [3]. This dataset,
reflective of real-world scenarios, incorporates transaction attributes such as time, type of
transaction, and amounts, serving as the foundational element for subsequent model
development [7]. Following data collection, a rigorous data preprocessing phase ensues,
addressing missing values, outliers, and inconsistencies to ensure a refined dataset [5].
This sets the stage for Exploratory Data Analysis, where patterns, trends, and correlations
between legitimate and fraudulent transactions are discerned [9].

A pivotal decision in the project lies in selecting a suitable machine learning algorithm for
fraud detection [5]. In this project, five different models based on distinct machine
learning algorithms are trained and compared based on their accuracy [8]. The chosen
model undergoes meticulous training with hyperparameters fine-tuned to optimize
performance [2]. Model evaluation is conducted on a separate testing dataset and
meticulously considers the business implications of false positives and false negatives
[11]. Upon successful evaluation, the model is deployed into the production environment,
becoming an integral part of the online payment system [9]. This phase marks the
beginning of continuous monitoring, where the model is updated regularly with new data
to ensure it remains effective in identifying evolving fraud patterns [1].

Despite advancements, traditional fraud detection methods often struggle to adapt to new
and evolving fraud tactics [11]. The aim of this project is to develop a fraud detection
system using machine learning techniques that classifies fraudulent transactions,
supporting threat mitigation practices at financial institutions [18]. However, challenges
arise due to the imbalanced nature of datasets, where legitimate transactions far
outnumber fraudulent ones, which can lead to model bias [19]. Although there is research
addressing class imbalance, there is limited work exploring anomaly detection algorithms
as a solution [12]. This project is conducted in collaboration with Deloitte Cyber to
leverage their expertise in understanding the security landscape for payments, the threats
involved, and the development of mitigation strategies [10]. Ultimately, the research
seeks to answer the question: To what extent can machine learning techniques support
fraud detection and payment security? [16].

2
2. Literature Survey
Fraud detection has become critical as the rise of online transactions has led to an increase
in fraudulent activities, especially in financial transactions such as credit card payments.
This survey explores various machine learning techniques applied to fraud detection,
drawing insights from previous studies on the subject.
Online fraud detection involves identifying fraudulent transactions while minimizing false
positives, ensuring legitimate transactions are not unnecessarily blocked. One of the
central challenges, highlighted in multiple works, is the imbalanced nature of transaction
data, where fraudulent transactions make up a very small portion of total transactions [5].
Effective fraud detection models must be capable of recognizing these rare events without
overwhelming the system with false alerts.
Supervised learning techniques such as logistic regression, decision trees, and support
vector machines (SVM) have been extensively used. These models rely on labeled datasets
to train algorithms on detecting fraudulent patterns [4]. However, they face challenges
when confronted with novel fraud patterns, which necessitates continuous retraining and
updating of the models [3].
Unsupervised learning, particularly anomaly detection techniques, can complement
supervised models by identifying outliers in transaction behaviour, which may indicate
fraud. These models are beneficial in identifying new fraud patterns, as they do not rely on
historical labels [7]. However, they tend to have higher false-positive rates, as they can
flag legitimate transactions that deviate from the norm [11].
Hybrid approaches that combine supervised and unsupervised methods have emerged as a
promising solution. These models aim to leverage the strengths of both techniques,
reducing false positives while maintaining the ability to detect new types of fraud. For
instance, combining clustering techniques with classification algorithms has been shown
to improve detection accuracy [9].

Real-time fraud detection has gained significant attention, especially with the
development of APIs that integrate machine learning models for immediate transaction
analysis. These systems allow banks and financial institutions to act swiftly in preventing
fraudulent activities [7]. Despite these advances, challenges remain in scaling these
systems for large volumes of data and addressing the computational demands of
continuous monitoring [13].

3
In conclusion, machine learning has revolutionized fraud detection, providing scalable
and adaptable solutions. However, challenges such as class imbalance, false positives,
and evolving fraud patterns require ongoing research and optimization [9]. Future work
could focus on improving real-time detection capabilities and developing more robust
hybrid models to handle complex fraud scenarios.

3. Motivation
The increasing incidence of online fraud, particularly in credit card transactions, is a
growing concern as digital transactions and e-commerce continue to expand globally [19].
Financial institutions and consumers are facing massive financial losses due to these
fraudulent activities, prompting a heightened demand for more effective fraud detection
systems [7]. Traditional systems, which primarily rely on rule-based methodologies, are
proving to be inadequate in the face of new and evolving fraud patterns [20]. These
systems are often rigid, generating high rates of false positives and false negatives, which
not only result in missed fraud but also cause unnecessary disruptions for legitimate
customers [4]. The limitations of these conventional systems underscore the need for
more advanced and adaptive approaches to fraud detection, capable of identifying
dynamic fraud tactics [8].

Machine learning has emerged as a promising solution to address the shortcomings of

traditional fraud detection systems [7]. By automating the detection of complex patterns
and anomalies in transaction data, machine learning models can significantly enhance the
speed and accuracy of fraud detection [9]. Real-time monitoring is another critical aspect,
as financial fraud can lead to significant losses within seconds [2]. This project aims to
develop a real-time fraud detection system using machine learning that allows for instant
identification and prevention of fraudulent activities [3]. The advancements in machine
learning not only improve fraud detection but also offer adaptive systems that can evolve
as new fraud tactics emerge, providing financial institutions with a robust defense
mechanism against the growing threat of online fraud [5]. Additionally, the adaptive
nature of machine learning models ensures that they can evolve alongside emerging fraud
tactics, maintaining their effectiveness even as cybercriminals change their strategies [9].
The advancements in machine learning not only improve fraud detection but also offer
adaptive systems that can evolve as new fraud tactics emerge, providing financial
institutions with a robust defense mechanism against the growing threat of online fraud,
thereby reducing the financial, operational, and reputational risks associated with these

4
incidents [10]. This ongoing evolution of fraud detection systems is crucial for
safeguarding the future of digital commerce and maintaining trust in financial institutions
[3].

4. Objectives
 To develop an online fraud detection system using machine learning
algorithms that identifies fraudulent transactions with high accuracy in real-
time.
 To minimize false positives and false negatives by designing models that
balance detection accuracy with business needs and operational costs.
 To address the challenge of imbalanced datasets by implementing
techniques such as oversampling, under-sampling, and anomaly detection
algorithms suited for rare-event detection.
 To compare and evaluate multiple machine learning models based on
performance metrics like precision, recall, F1-score, and accuracy to identify
the most effective solution for fraud detection.
 To ensure the fraud detection system adapts to evolving fraud patterns by
continuously updating the model with new transaction data.
 To collaborate with industry experts and incorporate current fraud
mitigation strategies to align the system with real-world security practices.
 To design the system for seamless integration into existing online payment
environments, providing real-time fraud monitoring and response capabilities.
 To utilize feature engineering techniques to extract and select the most
relevant transaction attributes that distinguish fraudulent behaviour from
legitimate activities.
 To implement hyperparameter tuning methods for optimizing the
performance of machine learning models, ensuring high detection rates while
maintaining computational efficiency.
 To explore and integrate advanced techniques such as ensemble learning or
hybrid models to enhance fraud detection accuracy by combining the strengths
of different algorithms.
 To evaluate the system’s robustness against adversarial attacks and ensure it
can withstand attempts by fraudsters to evade detection.

5
5. Architecture of the Proposed System

Figure 1 Architecture of Fraud Detection System

The architecture of the proposed system is designed to effectively detect fraudulent

activities by leveraging a combination of data processing, classification, and machine
learning techniques, as depicted in the diagram.

1. User Register/Login: The process begins when a user registers or logs into the
system. This ensures that all users are authenticated before initiating any
transactions.
2. Initiate Transaction: After the user logs in, they initiate a transaction. This
transaction is sent for further analysis to detect any potentially fraudulent activity.

6
 3. Process With Real-time Dataset: Once a transaction is initiated, it is processed
alongside a real-time dataset to provide immediate context for the classification
and prediction processes. The integration of real-time data ensures that the system
is up-to-date with the latest trends and patterns in transaction behaviors.

4. Pre-processed Dataset: Simultaneously, a pre-processed dataset is prepared. This

dataset is cleansed and standardized, ensuring it is suitable for analysis. Pre-
processing involves handling missing values, normalizing data, and removing
inconsistencies.
5. Attribute Selection: Attribute selection is performed on the pre-processed dataset
to identify the most relevant features for fraud detection. By selecting the key
attributes, the system enhances accuracy and reduces computational complexity
during classification.
6. Classification Techniques: The selected attributes and real-time data are then fed
into the classification techniques. This is where machine learning algorithms come
into play to determine whether the transaction is legitimate or fraudulent.
7. Implement Gradient Boosting & Implement Random Forest: Two major
classification techniques—Gradient Boosting and Random Forest—are
implemented in parallel. Gradient Boosting improves accuracy by combining
weak learners into a strong classifier, while Random Forest ensures robust
predictions by creating an ensemble of decision trees. The parallel implementation
of both techniques improves the reliability and performance of the fraud detection
process.
8. Fraud Detection: The final step, as shown in the diagram, is fraud detection. The
outputs from the Gradient Boosting and Random Forest algorithms are used to
make the final decision on whether a transaction is classified as fraudulent. The
system flags any suspicious transactions for further investigation.

By referring to the flow shown in the image, it is evident that the architecture efficiently
combines data processing and advanced machine learning techniques to detect fraud in
real-time with high accuracy. The use of both Gradient Boosting and Random Forest
strengthens the overall system performance, ensuring robust fraud detection capabilities.

7
6. Algorithm Description
 RANDOM FOREST
To distinguish between authentic and fraudulent transactions, the Random Forest
algorithm is utilized as a reliable and efficient instrument. To function, Random Forest
builds many decision trees, each of which is trained using a different subset of the
characteristics and data that are available. Together, these decision trees create an
ensemble, with each tree adding to the final classification.[5] The algorithm can recognize
suspicious activity depending on multiple transaction variables such transaction amount,
frequency, location, and user behaviour because it has learned patterns and relationships
within the transaction data during training. Random Forest's capacity to handle the
imbalanced nature of transaction datasets—where fraudulent cases are frequently greatly
outnumbered by genuine ones—is one of its main advantages for fraud detection.
This issue of class disparity can be successfully addressed by Random Forest by using
strategies like class weighting or modifying decision criteria, which will enhance its
accuracy in detecting fraudulent transactions. [7]. Moreover, high-dimensional data that is
frequently encountered in fraud detection applications is a good fit for Random Forest
models. Their real-time processing ability of vast amounts of transaction data makes them
perfect for online fraud detection systems that need to quickly identify any suspicious
activity. Furthermore, Random Forests guarantee dependable performance even in
dynamic transaction contexts since they are resistant to overfitting and noise in the data.
Random Forest also has the benefit of being interpretable. Through the examination of
feature important scores produced during model training, fraud analysts can acquire
valuable knowledge regarding the transaction attributes that most significantly affect that
have the greatest impact on the identification of fraudulent activity.[12]
 GRADIENT BOOSTING
One effective and popular method for spotting fraud is the Gradient Boosting algorithm.
Gradient Boosting, in contrast to conventional machine learning algorithms, creates an
ensemble of decision trees successively, with every new tree of decision trees in a

8
stepwise manner, with each new tree trying to improve on the mistakes of the one before
it.[3] By concentrating on the cases that were incorrectly classified in earlier iterations,
this iterative procedure enables the algorithm to progressively increase its forecast
accuracy. The methods are particularly effective in addressing imbalanced datasets, which
have a large proportion of genuine transactions compared to fraudulent transactions, in
the context of online transaction fraud detection. Also, it prioritizes the identification of
fraudulent transactions, improving overall performance by assigning greater weight to
misclassified occurrences. [2]
These models are skilled in identifying intricate patterns and relationships in transaction
information, allowing them to distinguish minute distinctions between authentic and
fraudulent activity. Gradient Boosting can efficiently leverage features like transaction
amount, frequency, location, and user behavior to spot suspicious activities. Furthermore,
the flexibility that Gradient Boosting algorithms provide with regard to model complexity
and parameter adjustment enables fraud detection systems to adjust to shifting fraud
patterns and dynamic threats. This flexibility is essential in dynamic online contexts
because fraud strategies are ever-changing. Moreover, Gradient Boosting offers feature
important insights that help fraud analysts determine which transaction The most crucial
elements for spotting fraudulent activity.[14]

7. Design Specification/Dataset Description

Figure 2 First Five Dataset

Ten thousand transaction details were used as training data among them we divide as type
of payment, amount, original name, old balance and new balance. One thousand datasets
were used for testing and real-time datasets were generated as transactions occurred. To
efficiently detect fraud transactions the real-time datasets are processed and compared
with the acquired dataset.[4]

The dataset also includes two key accounts involved in each transaction: nameOrig and
nameDest, which represent the originating and destination accounts, respectively.

9
Additionally, there are balance-related columns—oldbalanceOrg and newbalanceOrig
for the originating account, and oldbalanceDest and newbalanceDest for the destination
account—showing account balances before and after the transaction.[8]

8. Expected Outcomes
The primary expected outcome of this project is the development of a real-time fraud
detection system that leverages machine learning algorithms to classify and detect
fraudulent transactions in online payment systems. The system is expected to achieve the
following:

1. Real-time Fraud Detection: The system will be capable of analysing transactions

in real-time, identifying potentially fraudulent activities instantly, and alerting
administrators or blocking transactions before any damage occurs.

2. High Detection Accuracy: The system will compare various machine learning
algorithms (like Decision Trees, Random Forest, SVM, etc) to choose the most
accurate model, improving fraud detection rates while minimizing false positives
and false negatives.

3. Adaptive to Evolving Fraud Patterns: The system will incorporate machine

learning techniques to adapt to new and evolving fraud tactics, ensuring that it can
detect novel types of fraudulent activities without frequent manual updates or
retraining.

4. Reduced False Positives: Through the use of supervised and unsupervised

learning models, the system will aim to reduce the number of legitimate
transactions incorrectly flagged as fraudulent, improving the user experience for
genuine customers.

5. Handling Imbalanced Datasets: By implementing techniques like SMOTE and

under-sampling, the system will effectively manage imbalanced datasets where
fraudulent transactions make up a small portion, ensuring better detection of rare
fraud events.

10
6. Scalability: The system will be designed to handle large volumes of transactions
and can be deployed in high-traffic environments like e-commerce platforms,
payment gateways, and financial institutions.

7. Continuous Learning and Improvement: A feedback loop will be implemented

to continuous update the model based on new data and user feedback, making the
system more effective at detecting emerging fraud trends over time.

8. Improved Fraud Monitoring: The system will provide a user-friendly dashboard

or interface that displays flagged transactions, fraud probability scores, and the
reasoning behind the system's decisions, helping financial institutions to monitor
transactions efficiently.

9. Prevention of Financial Loss: With the timely detection and prevention of

fraudulent transactions, the system will help mitigate significant financial losses
for businesses and consumers.

10. Automated Detection Process: The system will automate the entire fraud
detection process, from data input to fraud identification, reducing the need for
human intervention and making the process more efficient.

11. Integration with Existing Systems: The fraud detection model can be integrated
into existing online payment systems via APIs, providing seamless real-time
protection without disrupting current operations.

12. Robust Evaluation Metrics: The system will be evaluated using key metrics like
accuracy, precision, recall, and F1-score to ensure it performs well under real-
world conditions and can distinguish between legitimate and fraudulent
transactions effectively.

13. Cost-Effective Solution: By automating the fraud detection process with machine
learning models, the system reduces the need for manual checks and minimizes
the operational costs for businesses.

14. Regulatory Compliance: The system will support compliance with financial
regulations related to fraud prevention, ensuring businesses stay within legal
frameworks while offering secure transaction processing.

11
 15. Enhanced Customer Trust: As a result of improved fraud detection and
prevention, the system will help build trust between financial institutions and their
customers, as they can rely on a secure, fraud-resistant transaction environment.

These outcomes will help significantly improve the monitoring, detection, and response
capabilities of financial institutions against online payment fraud.

References
[1]. Nghia Nguyen, Truc Duong, Tram Chau, Van-Ho Nguyen, Trang Trinh, Duy Tran, and Thanh Ho,
"A Proposed Model for Card Fraud Detection Based on CatBoost and Deep Neural Network,"
IEEE Access, University of Economics and Law, Vietnam National University, Ho Chi Minh City,
Vietnam.
[2]. Nadia Boutaher, Amina Elomri, Noreddine Abghour, Khalid Moussaid, and Mohamed Rida, "A
Review of Credit Card Fraud Detection Using Machine Learning Techniques," IEEE Access,
Hassan II University Casablanca, Morocco.
[3]. Cheng Wang, Songyao Chai, Hangyu Zhu, and Changjun Jiang, "CAeSaR: An Online Payment
Anti-Fraud Integration System With Decision Explainability," IEEE Transactions on Dependable
and Secure Computing, Senior Member, IEEE.
[4]. Suraya Nurain Kalid, Kok-Chin Khor, Keng-Hoong Ng, and Gee-Kok Tong, "Detecting Frauds
and Payment Defaults on Credit Card Data Inherited With Imbalanced Class Distribution and
Overlapping Class Problems: A Systematic Review," IEEE Access, Multimedia University,
Malaysia, Universiti Tunku Abdul Rahman, Malaysia.
[5]. Antonio Tudisco, Deborah Volpe, Giacomo Ranieri, Gianbiagio Curato, Davide Ricossa,
Mariagrazia Graziano, and Davide Corbelletto, "Evaluating the Computational Advantages of the
Variational Quantum Circuit Model in Financial Fraud Detection," IEEE Access, Politecnico di
Torino, Italy, Intesa Sanpaolo, Italy.
[6]. Seyedeh Khadijeh Hashemi, Seyedeh Leili Mirtaheri, and Sergio Greco, "Fraud Detection in
Banking Data by Machine Learning Techniques," IEEE Access, Kharazmi University, Iran,
University of Calabria, Italy.
[7]. Darshan Aladaktatti, Gagana P, Ashwini Kodipalli, and Shoaib Kamal, "Fraud Detection in Online
Payment Transaction Using Machine Learning Algorithms," IEEE Access, Global Academy of
Technology, Bangalore, India.
[8]. Fahdah A. Almarshad, Ghada Abdalaziz Gashgari, and Abdullah I. A. Alzahrani, "Generative
Adversarial Networks-Based Novel Approach for Fraud Detection for the European Cardholders
2013 Dataset," IEEE Access, Prince Sattam Bin Abdulaziz University, Saudi Arabia, University of
Jeddah, Saudi Arabia, Shaqra University, Saudi Arabia.
[9]. Abdulwahab Ali Almazroi and Nasir Ayub, "Online Payment Fraud Detection Model Using
Machine Learning Techniques," IEEE Access, University of Jeddah, Saudi Arabia, Air
University, Pakistan.
[10]. Cheng Wang and Hangyu Zhu, "Representing Fine-Grained Co-Occurrences for Behavior-Based
Fraud Detection in Online Payment Services," IEEE Access, Senior Member, IEEE.
[11]. Domenig, Thomas & Zvizdic, Ermin & Vanini, Paolo & Rossi, Sebastiano. (2022). Online
Payment Fraud: From Anomaly Detection to Risk Management.
[12]. Jack Nicholls, Aditya Kuppa, and Nhien-An Le-Khac, “Financial Cybercrime: A Comprehensive
Survey of Deep Learning Approaches to Tackle the Evolving Financial Crime Landscape” IEEE
Access, University College Dublin, Ireland.
[13]. Ranran Li, Zhaowei Liu, Yuanqing Ma, Dong Yang, and Shuaijie Sun, "Internet Financial Fraud
Detection Based on Graph Learning," IEEE Access, Graduate Student Member, IEEE.
[14]. Fuad A. Ghaleb, Faisal Saeed, Mohammed Al-Sarem, Sultan Noman Qasem, and Tawfik Al-
Hadhrami, "Ensemble Synthesized Minority Oversampling-Based Generative Adversarial
Networks and Random Forest Algorithm for Credit Card Fraud Detection," IEEE Access,

12
 University Teknologi Malaysia, Birmingham City University, Taibah University, Imam
Mohammad Ibn Saud Islamic University, and Nottingham Trent University.
[15]. Alarfaj, Fawaz & Malik, Iqra & Khan, Hikmat & Almusallam, Naif & Ramzan, Muhammad &
Ahmed, Muzamil. (2022). Credit Card Fraud Detection Using State-of-the-Art Machine Learning
and Deep Learning Algorithms. IEEE Access. 10. 1-1. 10.1109/ACCESS.2022.3166891.
[16]. Reem M. Own, Sameh A. Salem, and Amr E. Mohamed, "TCCFD: An Efficient Tree-based
Framework for Credit Card Fraud Detection," IEEE Access.
[17]. Nashwa Shaker Ragab, Omnia Elrashidy, Omar Adel, et al., “Fraud_Detection_ML: Machine
Learning Based on Online Payment Fraud Detection,” Journal of Computing and Communication,
vol. xx, February 2024. DOI: 10.21608/jocc.2024.339929.
[18]. Aditya Oza, "Fraud Detection Using Machine Learning," Stanford University.
[19]. M. Venkatesh, Bhukya Keerthi Bai, Budati Bhargavi, Chilukoti Manasa, and Dwarasala
Mokshitha, "Online Payment Fraud Detection Using Machine Learning," Vasireddy Venkatadri
Institute of Technology, Guntur, Andhra Pradesh, India.
[20]. E. Ileberi, Y. Sun, and Z. Wang, "Performance Evaluation of Machine Learning Methods for
Credit Card Fraud Detection Using SMOTE and AdaBoost," IEEE Access, vol. 9, pp. 165286-
165294, 2021.

Remarks/Comments to be Filled by the Department Project

Coordination Committee (DPCC)
Guide’s Remarks

Name of the Guide Signature

Panel Head Remarks

13
Name and Signature of the Panel Head

14