Explainable AI supported Evaluation and Comparison on Credit Card Fraud Detection Models

Abstract

This research investigates credit card fraud detection through the lens of machine learning and explainable AI techniques. We employed four distinct models: K-Nearest Neighbors (KNN), Random Forests, Extreme Gradient Boosting (XGBoost), and Logistic Regression. The SHapley Additive exPlanations (SHAP) method was used to enhance the interpretability of our models. Model performance was assessed using key metrics such as accuracy, precision, recall, and F1 score before and after feature selection. Notably, despite a decrease in some performance metrics post feature selection, high precision scores were maintained, underscoring the robustness of our models. Our findings lay the groundwork for future research in this field, highlighting the potential of a broader range of models, advanced explainable AI techniques, and innovative feature selection methods in the ongoing pursuit of robust and interpretable fraud detection systems.