Credit Card Fraud Detection: A Selection of A Few Machine Learning Algorithms

The increasing dependence on digital financial transactions has coincided with a rise in credit card fraud, which necessitates the development of advanced detection mechanisms that surpass traditional, static rule-based systems. The study undertakes a comparative analysis of six supervised machine learning algorithms, Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, k-Nearest Neighbours, and XGBoost, to detect fraud using a publicly available dataset. A central methodological focus was mitigating the extreme class imbalance, where fraudulent instances constitute less than 0.5% of the data, managed by applying the Synthetic Minority Over-sampling Technique (SMOTE). Ensemble methods, notably Random Forest and XGBoost, achieved superior and statistically significant results, nearing perfect classification, concluding that tree-based ensemble learners, when coupled with effective imbalance handling strategies, represent a highly dependable solution for modern credit card fraud detection.