Explainable AI (XAI) Analysis Using SHAP for Credit Card Fraud

Abstract

The increased use of credit cards in digital payment systems has also increased the risk of transaction fraud, which has led to financial losses and a decline in user confidence. Various machine learning approaches have been developed to automatically detect fraud, but most high-performance models are black-box in nature, making them difficult to explain and unsupportive of auditing and decision-making processes. This study aims to analyze the application of Explainable Artificial Intelligence (XAI) using the SHAP (SHapley Additive exPlanations) method in credit card fraud detection systems. An imbalanced credit card transaction dataset was used as experimental data, with two classification models, namely Logistic Regression as a baseline and Random Forest as an ensemble model. Performance evaluation was conducted using Precision, Recall, F1-score, and Average Precision (PR-AUC) metrics, which are more suitable for imbalanced data cases. The experimental results show that the Random Forest model performs better than Logistic Regression, especially in terms of Precision, F1-Score, and PR-AUC metrics. Explainability analysis using SHAP was performed to obtain global and local explanations for the model's decisions. Global explanations successfully identified the dominant features that influence fraud predictions, while local explanations provided an overview of the contribution of individual features to specific fraud transactions. The results of this study show that the application of SHAP can improve the transparency and clarity of fraud detection model decisions without sacrificing prediction performance, thereby potentially supporting the development of a more reliable and easily audited fraud detection system.