Detecting Medical Insurance Fraud Using a Heterogeneous Information Network with a Multi-behavior Pattern

Abstract

Medical insurance frauds impose huge losses on the medical insurance funds. Hence, efficient detection of medical insurance frauds is essential for sustainable medical insurance funds and affects the quality and affordability of health care. The existing detection methods determine fraudulent cases by statistical analyses of the patients’ characteristics using graph neural network models. Nevertheless, such methods ignore the potential relationship between the graph nodes which represents the degree of similarity between the patients’ characteristics. To address this issue, we propose a medical insurance fraud detection model based on a heterogeneous information network with a multi- behavior pattern, namely $$\mathrm{BPGAN}$$ . In our proposed approach, we analyze the characteristics of different behavior patterns among patients and design an attentional mechanism for embedded learning of patient nodes in the medical insurance data network. It enables capturing potential relationships between the patients’ nodes and obtaining the corresponding weight coefficients of different behavior patterns in the process of embedded learning. Experimental results on real medical insurance data sets confirm that the proposed method effectively identifies medical insurance fraudsters and overperforms existing fraud detection methods.