A Privacy-Aware Explainable Artificial Intelligence Framework for Financial Fraud Detection Using Graph Neural Networks

Abstract

The rapid growth of digital payment systems and online banking has significantly increased the occurrence of sophisticated financial fraud. Traditional fraud detection methods often fail to identify complex fraud patterns while lacking transparency in decision-making. This paper proposes a privacy-aware Explainable Artificial Intelligence (XAI) framework that integrates Graph Neural Networks (GNNs) with attention mechanisms to detect fraudulent financial transactions. Financial entities are represented as graph structures, enabling the identification of hidden relationships among users, devices, and transactions. SHAP and GNNExplainer techniques are incorporated to provide interpretable predictions for financial analysts. Privacy-preserving mechanisms, including federated learning and differential privacy, ensure secure model training without exposing sensitive financial information. Experimental results demonstrate improved fraud detection accuracy, reduced false-positive rates, and enhanced interpretability compared with conventional machine learning approaches. The proposed framework provides a trustworthy and scalable solution for secure financial analytics.