Published Paper Details:

In today's digital financial systems, fraud detection models must balance high accuracy with interpretability to meet regulatory and operational demands. This paper presents a novel Explainable AI (XAI) framework that combines SHapley Additive exPlanations (SHAP) with unsupervised clustering to enhance transparency and uncover latent fraud patterns. We train a Random Forest model on a real-world credit card dataset (284,807 transactions, 492 fraudulent) and compute SHAP values to quantify feature contributions. By applying t-SNE dimensionality reduction and k-means clustering to SHAP explanations, we identify three distinct fraud typologies (e.g., high-amount frauds, identity theft indicators) that traditional methods overlook. Our approach achieves 93% precision for fraud detection while providing auditable, global interpretations of model behavior. The integration of SHAP and clustering enables financial institutions to segment risks, adapt to emerging fraud strategies, and comply with regulations like GDPR. This work bridges the gap between model explain ability and actionable fraud analytics, offering a scalable solution for ethical AI adoption in finance.