Ensemble Methods and Emerging Paradigms in Credit Card Fraud Detection: A Comparative Study

Credit card fraud is a major problem for financial institutions worldwide, with losses reaching over \$32 billion each year. In this paper, we present a detailed comparison of machine learning methods for detecting credit card fraud. We tested 52 different approaches including traditional methods, ensemble techniques, deep learning, and newer algorithms using the well-known MLG-ULB dataset. Our results show that ensemble methods, especially Random Forest, work best overall with accuracy up to 99.98\% and F1-scores of 0.87. We looked at important factors like model accuracy, how fast they run, if we can understand their decisions, and if they work in real systems. The results give us useful information about handling imbalanced data, choosing features, and what problems we might face when using these models in practice. This work helps people choose the right algorithm for real fraud detection systems while thinking about rules and regulations and how big the system needs to be.