Enhancing Medicare Fraud Detection with a CNN–Transformer– XGBoost Framework and Explainable AI

Abstract

Medicare fraud is a major challenge in the healthcare sector, resulting in financial losses, inefficient utilization of healthcare resources, and increased administrative burden. Traditional rulebased systems and manual auditing methods are often time-consuming and less effective in identifying complex and evolving fraudulent claim patterns. This paper proposes a hybrid Medicare fraud detection framework that integrates Convolutional Neural Network (CNN), Transformer, XGBoost, and Explainable Artificial Intelligence (XAI). The CNN model extracts important local feature representations from healthcare claim data, while the Transformer captures contextual and long-range relationships among claim attributes using self-attention mechanisms. The extracted representations are then provided to the XGBoost classifier to distinguish fraudulent and legitimate claims. To improve transparency, XAI techniques such as SHAP and LIME are incorporated to identify the features contributing to individual predictions. The proposed framework is evaluated using accuracy, precision, recall, F1-score, ROC-AUC, and confusion matrix measures. The project reports approximately 99% precision and 98% recall, demonstrating the potential of the hybrid approach for accurate and reliable fraud detection. The integration of XAI further supports healthcare analysts and fraud investigators by providing understandable explanations for suspicious predictions. Overall, the proposed framework provides an accurate, scalable, and transparent approach for intelligent Medicare fraud detection. Keywords: Medicare Fraud Detection, Convolutional Neural Network (CNN), Transformer, XGBoost, Explainable Artificial Intelligence (XAI), Deep Learning, Healthcare Analytics, Fraud Detection.