Interpretable Credit Card Fraud Detection Using Machine Learning Leveraging SHAP

More and more companies are taking their services online as a result of how widely accessible the internet is. Besides, because of the growth of E-commerce websites, both individuals and businesses that deal in finances are more dependent on internet administrations to handle their business. Since more and more people are using online banking and making purchases online, credit card fraud has increased. These days, fraudsters use a variety of cutting-edge ways to thwart the methodical operation of the fraud detection system that is already in place. In this research, we have implemented six different machine learning (ML) models to develop an efficient fraud detection system. The techniques in our research include Logistic Regression, SVM, Random Forest, AdaBoost, and AdaBoost with Random Forest and Logistic Regression as the base estimators. Furthermore, to illustrate the classification results, we employed SHapley Additive exPlanations (SHAP) to demonstrate the explainability of the two top-performing models. This research shows that the Random Forest and AdaBoost with Random Forest as a base estimator models, are the most effective algorithm with a test accuracy of 99.999% each. Besides, these two models have performed very well in terms of Precision, Recall, and F1-Score than the other ML models.