Credit Card Fraud Detection Using
Machine Learning Techniques
Project report submitted to
Indian Institute of Information Technology, Nagpur,
in partial fulfillment of the requirements for the award of the degree
of
Bachelor of Technology
In Electronics and Communication
Engineering
By
Sandeep Manelli (BT21ECE105)
Umesh Chandra (BT21ECE117)
Narahari Reddy (BT21ECE127)
Uday Singh (BT21ECE131)
Under the Guidance of
Dr. Parul Sahare & Dr. Nikhil Agarwal
Assistant Professors, ECE Department
Indian Institute of Information Technology, Nagpur
Indian Institute of Information Technology, Nagpur
Nagpur 441108 (India)
2021-2025
2
Declaration
We, the undersigned, hereby declare that the project work titled “Credit Card Fraud
Detection using Machine Learning Techniques” is an original work carried out by us
in the Department of Electronics and Communication Engineering, Indian Institute
of Information Technology, Nagpur, under the guidance of Dr. Parul Sahare and Dr.
Nikhil Agarwal.
We affirm that the work is authentic, has not been submitted earlier for any degree or
diploma, and adheres to the academic standards and policies of the institute.
Team Members:
Sr. No. Name Enrollment No. Signature
1 Sandeep Manelli BT21ECE105
2 Narahari Reddy BT21ECE127
3 Uday Singh BT21ECE131
4 Umesh Chandra BT21ECE117
Date:
Guidance Acknowledgment:
Dr. Parul Sahare & Dr. Nikhil Agarwal
Assistant Professors, Department of ECE
Indian Institute of Information Technology, Nagpur
We, the undersigned, understand that plagiarism is defined as any one or combination of
the following:
1. Uncredited verbatim copying of individual sentences, paragraphs, or illustrations
(such as graphs, diagrams, etc.) from any source, published or unpublished, includ-
ing the internet.
2. Uncredited improper paraphrasing of pages or paragraphs (changing a few words
or phrases or rearranging the original sentence order).
3. Credited verbatim copying of a major portion of a paper (or thesis chapter) with-
out clear delineation of who did or wrote what. For example, (Source: IEEE, the
institute, Dec. 2004).
I affirm that no portion of this work can be considered as plagiarism, and I take full
respon- sibility if such a complaint occurs. I fully understand that the guide of the thesis
may not be in a position to check for the possibility of such incidences of plagiarism in
this body of work.
Signatures of Team Members
Uday Singh Narahari Reddy
Sandeep Manelli Umesh Chandra
2
Certificate
This is to certify that the project titled “Credit Card Fraud Detection Using Machine
Learning Techniques”, submitted by
Uday Singh (BT21ECE131), Sandeep Manelli (BT21ECE105), Narahari Reddy
(BT21ECE127), and Umesh Chandra (BT21ECE117),
in partial fulfillment of the requirements for the award of the degree of Bachelor of
Technology in Electronics and Communication Engineering, IIIT Nagpur, is a record
of bona fide work carried out under our guidance. The work is comprehensive, complete,
and fit for final evaluation.
Date: Dr. Parul Sahare
Assistant Professor, ECE
IIIT Nagpur
Dr. Nikhil Agarwal
Assistant Professor, ECE
IIIT Nagpur
Dr. Harsh Goud
Head of the Department, ECE IIIT, Nagpur
3
Acknowledgments
We take this opportunity to express our sincere gratitude to those who have supported
and guided us throughout the duration of this project.
First and foremost, we extend our deepest thanks to our project mentors, Dr.
Parul Sahare and Dr. Nikhil Agarwal, Assistant Professors in the Department of
Electronics and Communication Engineering, for their invaluable guidance,
encouragement, and in- sights that were pivotal in shaping this work. Their expertise and
constructive feedback have provided us with clarity and direction, enabling us to achieve
the objectives of this project.
We are also grateful to the Indian Institute of Information Technology, Nag-
pur, for fostering an environment that emphasizes academic rigor and innovation. The
resources and infrastructure provided have significantly contributed to the completion of
our work.
A special note of thanks to Dr. Prem Lal Patel, Director of IIIT Nagpur, for
leading an institution that supports interdisciplinary learning and technical excellence.
His leadership has been instrumental in cultivating a culture of research and
development.
We wish to extend our gratitude to the faculty members of the Department of
Elec- tronics and Communication Engineering for their continuous support and for
sharing their expertise during our academic journey.
Finally, we would like to express our heartfelt appreciation to our families and
friends for their unwavering encouragement and support, which provided us with the mo-
tivation to persevere and complete this project.
The Team
4
Umesh Chandra, Uday Singh, Sandeep Manelli, Narahari Reddy
5
Abstract
This project focuses on developing a sophisticated AI-driven system aimed at enhancing
the security and reliability of credit card transactions by detecting fraudulent activities
with high accuracy and efficiency. The system leverages advanced machine learning
models, including Logistic Regression (LR), Random Forest (RF), Decision Trees (DT),
and K-Nearest Neighbors (KNN), to analyze transaction patterns, identify anomalies, and
differentiate between legitimate and fraudulent transactions.
By using Logistic Regression, the system can establish a probabilistic baseline to classify
transactions based on linear relationships between features. The Random Forest model
enhances detection by combining the predictions of multiple decision trees, ensuring
robustness and reducing overfitting. Decision Trees contribute by offering clear
interpretability, enabling stakeholders to understand the reasoning behind fraud detection.
The inclusion of K-Nearest Neighbors allows the system to classify transactions based on
proximity to similar historical data points, further improving detection in non-linear and
complex scenarios.
The system is designed to process transactional data in real-time, ensuring swift
responses to potentially fraudulent activities. Through natural language understanding
(NLU), users can receive clear, contextual explanations for flagged transactions,
improving transparency and trust. Additionally, detailed dashboards and reports provide
stakeholders with actionable insights into transaction trends and fraud patterns.
Implemented as a web-based application, the solution is accessible on a variety of
devices, enabling financial institutions, merchants, and users to benefit from its
functionality without requiring extensive technical setups. By combining cutting-edge AI
with practical usability, this system aims to significantly enhance the security of credit
card transactions, reduce financial losses due to fraud, and promote trust in digital
payment systems
6
TABLE OF CONTENTS
Sr. No. Topic Page
1 Introduction
1.1. Introduction to problem 9-10
1.2. Objective
1.3. Applications
2 Literature Review 11-12
3 Dataset Preparation
3.1. Data Preprocessing 13-15
3.2. Handling Imbalanced Classes
4 Methodology
4.1. Credit Card Fraud Detection Architecture
4.1.1. Data Preprocessing Module 16-18
4.1.2. Feature Extraction and Selection Module
4.1.3. Model Training and Evaluation Module
4.2. Steps in Fraud Detection
4.2.1. Two-Stage Training Approach
4.2.2. Feature Conditioning Strategy 18-22
4.2.3. Model Comparison and Ensemble Methods
4.2.4. Performance Optimization Objective
5 Performance Evaluation
5.1. Accuracy
5.2. Precision Score 23-29
5.3. Recall Score
5.4. F1-Score
5.5. Results and analysis
5.6. Limitations
5.7. Applications
6 Conclusion
6.1. Future Scope
7 References
7
CHAPTER 1
INTRODUCTION
Credit card fraud detection involves creating a system that can accurately identify
fraudulent transactions while minimizing false positives. This requires tackling
challenges such as feature extraction, handling imbalanced data, and detecting
anomalies in spending patterns. By using machine learning techniques like supervised
learning and anomaly detection, the goal is to develop a model that differentiates
legitimate transactions from fraudulent ones, ensuring real-time detection and
continuous improvement over time for secure financial protection.
1.1 Introduction:
Credit card fraud refers to unauthorized or deceptive use of credit card information to
obtain goods, services, or funds, causing substantial financial losses and eroding trust in
digital payment systems.
The complexity of credit card fraud arises from its evolving nature, where fraudsters
constantly devise new techniques to bypass traditional security measures. Common types
of fraud include identity theft, account takeovers, counterfeit card use, and online
payment fraud. With millions of transactions occurring daily, identifying fraudulent
activity amidst legitimate transactions becomes a daunting task.
To address this challenge, Credit Card Fraud Detection leverages advancements in data
analysis and machine learning to analyze vast amounts of transaction data in real-time.
These systems aim to identify patterns, detect anomalies, and flag suspicious activities.
By employing techniques such as Logistic Regression, Random Forests, Decision
Trees, and K-Nearest Neighbors, fraud detection systems can effectively distinguish
between legitimate and fraudulent transactions.
8
1.2. Objective
The project aims to develop a system that accurately detects fraudulent credit card
transactions by analyzing transaction data. It focuses on modeling the behavior patterns
of legitimate transactions and identifying anomalies that signal potential fraud. By
leveraging machine learning techniques, the system will synthesize patterns from
historical transaction data to distinguish between valid and fraudulent activities. This
approach bridges the gap between transaction data and fraud detection, offering
applications in financial security, real-time fraud prevention, and protection for both
consumers and financial institutions.
1.3. Applications
This model will assist financial institutions in detecting fraudulent transactions by
analyzing spending patterns and identifying discrepancies, potentially preventing
significant financial losses. By identifying unusual activity, it helps protect consumers
and businesses from credit card fraud, contributing to more secure online and offline
transactions.
Retailers and e-commerce platforms can leverage the system to identify and block
fraudulent purchases in real-time, ensuring a safe shopping experience for customers. By
analyzing transaction data, the model can identify patterns of fraud that may be difficult
for traditional methods to detect, helping to maintain trust in online shopping.
Banks and payment processors will improve fraud detection efficiency by automating
the process of reviewing transactions. By quickly identifying high-risk transactions, the
model can enable faster response times, reducing the need for manual reviews and
enhancing overall customer service.
9
CHAPTER 2
LITERATURE REVIEW
This paper proposes a method for detecting credit card fraud using a multi-layered
machine learning approach. The system incorporates various techniques, including
decision trees, support vector machines (SVM), and deep learning models, to identify
fraudulent transactions in real-time. By analyzing transaction data and incorporating
features such as spending patterns, location, and device information, the approach
significantly improves fraud detection accuracy. While computationally intensive, the
method demonstrates strong potential for use in financial systems, offering an effective
solution for both reducing fraud and maintaining operational efficiency in payment
processing systems.
In the field of credit card fraud detection, the challenge often lies in detecting subtle
anomalies in transaction data, especially given the imbalance between fraudulent and
legitimate transactions. To address this, Chen et al proposed an innovative approach
where transaction data is decomposed into key components such as merchant,
transaction amount, time of day, and geographic location. Each of these features is then
processed through separate models, such as autoencoders, to detect unusual patterns.
By applying K-nearest neighbor search techniques to combine the results of these
models, the system is able to predict fraudulent activity with higher accuracy, even in
cases of limited or imprecise transaction data.
A significant limitation of traditional fraud detection systems is their inability to
effectively learn the dependencies between various transaction features. Lei et al
proposed an approach that uses a multi-scale architecture and self-attention
mechanisms to address this issue.
10
11
