Trustworthy and Interpretable AI for Robust Fraud Detection in Financial Transactions

Financial fraud poses a critical threat to global economies, necessitating robust and transparent detection mechanisms. Traditional machine learning models achieve high predictive accuracy but often lack interpretability, limiting their adoption in high-stakes financial decision-making. This work introduces a novel Interpretable AI (IAI) framework that enhances fraud detection by integrating Explainable Boosting Machines (EBMs), SHAP (SHapley Additive exPlanations), and causal inference models. Our approach ensures high detection performance while providing clear, human-understandable explanations for each decision, fostering regulatory compliance and trust among financial institutions. Secondly, we use Federated Learning (FL) to facilitate privacy-preserving multi-party collaborative fraud detection among institutions without exchanging sensitive information. Experimental testing on actual transaction datasets illustrates 20–30% better recall of fraud compared to traditional black-box models, with retaining computational efficiency. We also test the performance of our model using adversarial evaluation to ensure robustness against complex fraud patterns. The suggested IAI framework is compatible with new regulatory needs and ethical AI standards, making it an affordable and scalable solution for future-proofed fraud prevention systems.