Fraud Guard: A Comprehensive Comparative Analysis of Machine Learning Approaches to Enhance Credit Card Fraud Detection

Abstract

The COVID-19 pandemic has constrained people's mobility, prompting a surge in reliance on online services due to challenges in offline purchasing. Machine learning (ML) methods have played a crucial role in advancing classification and prediction techniques across various domains. In the realm of Credit Card Fraud Detection, the significance of ML is particularly pronounced. These methods harness the power of data-driven algorithms to distinguish between legitimate and fraudulent transactions, contributing significantly to the enhancement of security measures in financial transactions. The dynamic and adaptive nature of ML allows for the continuous evolution of fraud detection systems, ensuring a proactive approach to safeguarding against emerging threats in the credit card landscape. With this shift, credit card fraud has become a significant concern within the domain of internet-based transactions. Hence, there is a pressing demand to devise an optimal machine learning method for preventing fraudulent credit card transactions. The study employed four resampling techniques (CNN, AllKNN, SMOTE, and SVMSM ) and three machine learning approaches (XGBoost , CatBoost, and RF) for analysing credit card fraud datasets with the aim of detection. These findings demonstrated that integrating AllKNN as an undersampling technique and CatBoost as a classifier  are achieving superior results across the evaluated methods. The accuracy, precision, recall, and f1-score were 99.9%, 95.9%, 80%, and 87.4%, respectively.

Keywords: Unbalanced data, machine learning techniques, fraud detection, and credit card fraud.

DOI: 10.7176/JIEA/14-2-02

Publication date:March 31^st 2024