Enhancing Machine Learning for Credit Card Fraud Detection: A Comparison of Supervised Models

In the digital age, fraud involving credit cards has become a serious concern that impacts lots of people and companies worldwide. Reliable and intelligent fraud detection systems are becoming more and more important as e-commerce and online financial transactions grow in popularity. Conventional rule-based algorithms sometimes struggle with big, unbalanced datasets and are unable to identify new fraud tendencies. Using anonymized real-world data, this study explores the exploitation of Machine Learning models so that efficiently fraudulent credit card transactions could be identified. In this work three popular supervised learning models that are Random Forest, XGBoost, and Logistic Regression were used. Because of the extreme imbalance of the information, just 0.17% of transactions were deemed fraudulent, pre-processing techniques like normalization and the Synthetic Minority Oversampling Technique (SMOTE) is being utilized for balancing group in order to overcome this. Key performance indicators like Positive Predictive Value, sensitivity, f-measures, accuracy, and the region beneath the ROC curve (AUC-ROC) are used for evaluating the models’ performance. With an AUC of 0.94, XGBoost outperforms the other models examined, demonstrating its resilience to differentiate between real and fake transactions. Findings suggest ensemble learning models XGBoost are very good at detecting fraud and can be included into real-time financial monitoring systems to improve security.