Using Machine Learning Algorithms for Advanced Credit Card Fraud Detection

Abstract

Credit card fraud (CCF( is a serious problem that impacts a number of industries, including banking, e-commerce, insurance, finance, commercial entities, and private citizens. More sophisticated technologiesare needed for effective detection as fraudulent activities become more complex. Machine learning (ML)approaches have demonstrated high performance in CCF identification. In this study, we explore six MLtechniques: CatBoost, Hist Gradient Boosting, Decision Tree (DT), LightGBM (LGBM), ANN, and XGBoost(XGBC). To represent the data, identify its key characteristics, and compare their performance to determinethe most successful fraud detection algorithm. Among these algorithms, the CatBoost algorithm, known forits high accuracy and superior performance, stands out. The proposed approach was based on the Kaggledataset, and the results showed that CatBoost achieved the highest performance with a lower error ratecompared to other models. This study also discusses the challenges related to implementing these advancedtechniques, including the requirement for sizable, high-quality datasets and significant computationalresources for model training and deployment. Key performance measures were used to evaluate thetechniques' effectiveness, and CatBoost achieved a test accuracy of 0.9966, outperforming Hist GradientBoosting, DT, LGBM, XGBC and ANN.