Credit Card Fraud Detection with XAI: Improving Interpretability and Trust

Fraud detection is crucial for financial institutions to protect clients' funds and maintain trust in online payment systems. However, traditional machine learning models often lack interpretability due to advanced fraud methods. This study proposes a framework that integrates advanced machine learning algorithms with XAI to improve the interpretability of fraud detection models. The framework explains model predictions clearly, helping stakeholders understand and verify system decisions. The Kaggle dataset is pre-processed, i.e., classes are balanced before implementing the ML techniques. The pre-processed Kaggle dataset is used to implement various ML algorithms, with the Local Interpretable Model-agnostic Explanations (LIME) method used during the prediction phase. Decision tree and Random Forest outperform other ML techniques, achieving 100% accuracy. This method clarifies the decision- making process, allowing stakeholders like financial analysts, lawmakers, and consumers to trust and confirm system results.