Transformer Model for Fraud Detection in Medical Insurance Claims

Healthcare fraud is a major problem that leads to wasted money and decreased efficiency in medical insurance systems. Traditional fraud-detection methods based on fixed rules struggle to adapt to new scam techniques and often misclassify valid cases as fraudulent. In this article, we put forward a Transformer-Based Anomaly Detection model to enhance fraud detection in medical insurance claims. The model uses self-attention, a feature of transformers, to examine the Medical Malpractice Insurance Dataset and spot unusual patterns by evaluating the reconstruction error thresholds. Developed with PyTorch Lightning, the system’s effectiveness is evaluated through metrics that include AUC-ROC, Silhouette Score, and Confusion Matrix. Our results indicate that this model successfully identifies fraudulent claims while significantly lowering the number of false positives. By providing a scalable and adaptable solution, this approach overcomes the limitations of traditional methods, strengthens the fraud detection framework in the insurance industry, and ultimately protects financial and economic efficiency.