Health insurance fraud detection based on multi-channel heterogeneous graph structure learning

Area of Science:

• Computer Science
• Data Science
• Health Informatics

Background:

• Health insurance fraud is increasing, undermining system fairness and sustainability.
• Traditional fraud detection methods struggle with complex, evolving data and fraud tactics.
• There is a critical need for advanced, adaptable analytics to detect health insurance fraud effectively.

Purpose of the Study:

• To introduce the Multi-channel Heterogeneous Graph Structured Learning (MHGSL) method for health insurance fraud detection.
• To leverage graph structure learning and deep learning for improved fraud identification.
• To enhance the accuracy and efficiency of detecting fraudulent activities in health insurance data.

Main Methods:

• Constructing a heterogeneous graph from diverse health insurance entities (patients, departments, medicines).
• Employing graph structure learning to extract topological, feature, and semantic information.
• Utilizing deep learning (heterogeneous graph neural networks, graph convolutional neural networks) for multi-channel information fusion and anomaly detection.

Main Results:

• MHGSL demonstrated high accuracy in detecting potential health insurance fraud, outperforming existing methods.
• The method effectively and rapidly identifies patients exhibiting fraudulent behaviors.
• Experiments confirmed the significant contribution of multi-channel heterogeneous graph structure learning to fraud detection efficacy.

Conclusions:

• MHGSL offers a promising solution for detecting health insurance fraud, enhancing system fairness and sustainability.
• The approach effectively addresses the limitations of traditional fraud detection methods.
• Future research should explore incorporating semantic information between patients and various entities for further improvements.