Online Payments Fraud Detection Using Machine Learning Techniques
Hend Abdelbakey Atia, Magdy Aboul-Ela, Christina Albert Reyad, Nancy Awadallah Awad · 2024
With the rapid growth of online transactions and e-commerce, concerns about the security of online payment systems have increased. Additionally, the rising incidence of fraudulent activity imposes higher costs on bank transactions, posing a significant threat to businesses, especially credit card fraud. To effectively combat deception, it is crucial to understand the strategies employed by criminals targeting credit card transactions. Consequently, there has been considerable interest in identifying fraudulent operations. In this study, we propose a model based on data mining techniques and machine learning algorithms that outperforms rule-based algorithms for online payment fraud detection. We compare the performance of Categorical Boosting (CatBoost), eXtreme Gradient Boosting (XGBoost), and Light Gradient-Boosting Machine (LightGBM - LGBM) models. Experimental results demonstrate that the LightGBM model achieves the highest performance metrics. With an accuracy of 0.998, the LightGBM model accurately distinguishes between fraudulent and valid transactions. It also exhibits a high sensitivity (recall) score of 0.989, indicating its ability to detect fraudulent cases. The precision score of 0.346 suggests a possibility of false positives. The F1-score, which provides a balanced measure of recall and precision, is 0.513, representing a fair compromise. The Area Under the Curve (AUC-ROC) score of 0.9933 reflects the model's discriminative capability. The Matthews correlation coefficient (MCC) shows a moderate agreement of 0.584 between actual labels and predictions.