Application of Machine Learning Algorithms in Detecting Credit Card Fraud: A Comparative Analysis

Abstract

Credit card transaction have grown increasingly prevalent in the digital era, and along with them, so have incidents of associated fraud. Hence, identification and prevention of such frauds are critically crucial. Machine learning algorithms are predominantly employed in the realm of credit fraud detection. According to current literature, class imbalance of data, a great disparity in ratio between normal and fraudulent transactions, could severely affect the result in detection. In this paper, a combination of imbalanced classification methods, specifically the Synthetic Minority Random Oversampling Technique (SMOTE) and Under-sampling, is utilized to harmonize the dataset. Some popular machine learning algorithms are applied to detect frauds are compared and analyzed, including Logistic Regression, Decision Tree, Random Forest and XG Boost. The accuracy, precision, recall, F-1 score and Area Under Curve (AUC) of each algorithm are used as metrics of performance evaluation. The research findings indicated that among the four models tested, XG Boost, when coupled with balanced data yielded overall optimal results for classifying fraudulent activities.