A Comprehensive Analysis of Credit Card Fraud Detection Using Hybrid Oversampling and Machine Learning Models

Abstract

Detecting credit card fraud is a challenging problem due to heavy class imbalance inherent in real world transaction which only a small fraction of observation is classed as fraudulent. This imbalanced dataset greatly reduces the power of classical machine learning and deep learning models, as the performance accuracy of these models will be high but their fraud detection ability will be very low. This paper gives a systematic review on imbalance handling methods utilized in the credit card fraud detection with an emphasis on confusion-matrix–based performance evaluation. All the data-level techniques like SMOTE and its varieties, hybrid resampling techniques like SMOTE-ENN, SMOTE-Tomek and algorithm level techniques have been reviewed systematically. The literature shows performance trends that can be analyzed in terms of recall, precision, F measure, and false positive behavior. As suggested by the review, regularly, hybrid resampling approaches achieve a more balanced detection performance by improving fraud recall while managing false alarms. The findings offer actionable insights for selecting effective imbalance reduction plans in fraud detection systems.