Healthcare Fraud Detection Based on ChatGPT-Guided Method

This paper proposes a ChatGPT-guided method for healthcare fraud detection. By using ChatGPT-4.5 as the teacher model to generate fraud probability labels, and training a more efficient LightGBM student model, the approach achieves an accuracy of 88.47%ˈcompared with the traditional detection model, it has increased the accuracy rate by 15.3%.To address challenges such as data imbalance and privacy concerns, the proposed method provides a scalable and effective solution for cross-institutional fraud detection.