Due to issues such as cognitive differences and blurred boundaries, "three treatment fees" (expenses incurred from rehabilitation, physiotherapy, and traditional Chinese medicine treatments) have been maladies affecting the rational use of medical insurance funds and regulatory enforcement. The "three treatment fees" supervision model is a regulatory tool developed to address this issue, based on unsupervised self-learning for precise prediction and discovery of the unknown. By employing methods such as data preprocessing, multidimensional clustering, and suspicious physician cluster locking, the model is constructed to achieve periodic warning, real-time monitoring, and normalized handling. This model has been applied in the medical insurance supervision of a city, which can effectively identify potential suspicious physicians based on model alerts. Through inter-departmental cooperation, the model can effectively guide medical institutions and physicians to standardize their diagnosis and treatment behaviors using methods such as supervision and inspection, administrative discussions, and "driver's license" style scoring.