Loading...
A Machine Learning-Based Approach for Credit Card Fraud Detection
Salama, Fatma
Salama, Fatma
Citations
Altmetric:
Type
Supervisor
Date
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
This report addresses the critical challenge of credit card fraud detection
in highly imbalanced transaction datasets. Traditional rule-based systems
su!er from poor adaptability and high false-positive rates, while anomaly
detection models often struggle with precision in extreme imbalance scenar ios.
This research proposes and evaluates a fraud detection framework that
applies comprehensive preprocessing techniques and SMOTE oversampling
to handle class imbalance, followed by a systematic comparison of supervised
learning (XGBoost) with anomaly detection models (One-Class SVM and
Isolation Forest) on the publicly available European credit card transaction
dataset.
Experimental results demonstrate that applying SMOTE significantly im proves fraud detection performance across all models, particularly in terms
of recall for the minority fraud class. Among the evaluated models, XGBoost
achieves the best balance between fraud detection accuracy and false positive
rate, outperforming anomaly detection approaches in overall e!ectiveness.
The findings highlight the importance of combining appropriate preprocess ing techniques with supervised learning methods when dealing with highly
imbalanced transaction data.
This project provides a structured and reproducible framework for o"ine
credit card fraud detection and o!ers practical insights into model selection
and imbalance handling strategies in financial transaction analysis