Credit Card Fraud Detection Automation System Using Logistic Regression and XG Boost

Abstract

Credit card fraud poses a significant challenge to financial institutions, merchants, and customers, resulting in substantial financial losses and reduced consumer confidence. The rapid growth of digital payment systems has increased the volume and complexity of transactions, making automated and reliable fraud detection essential. This study proposes an automated machine learning framework for credit card fraud detection that addresses the severe class imbalance inherent in fraud datasets through Random Under-Sampling and Synthetic Minority Over-sampling Technique (SMOTE). The framework evaluates multiple approaches, including Logistic Regression, Random Forest, XGBoost, and an Autoencoder-based anomaly detection model, to provide a comprehensive comparison of supervised and unsupervised learning techniques. Unlike conventional approaches that primarily rely on accuracy, the proposed framework evaluates model performance using Precision, Recall, F1-score, ROC-AUC, and Precision-Recall AUC (PR-AUC), providing a more meaningful assessment of fraud detection capability. An initial Logistic Regression experiment using data preprocessing, exploratory analysis, statistical validation, and under-sampling achieved an accuracy of 93.91%. The extended framework aims to identify models that effectively balance the detection of fraudulent transactions with the reduction of false alarms. The proposed approach provides a scalable foundation for real-time transaction monitoring and can support adaptive fraud detection systems capable of responding to evolving fraud patterns.