Enhanced Medicare Fraud Detection Using Graph Convolutional Networks

This paper explores the applications of Graph neural networks (GNN) for enhancing Medicare fraud detection. Graph convolutional network (GCN) is a type of graph neural network. Governments and insurance companies are continuously adapting new technologies to detect and prevent fraud activities and trying to minimize financial losses and improve services because every year they lose billions of dollars due to Medicare fraud. Machine learning algorithms fail to analyze the graph data structure but Graph neural networks are good at analyzing the complex relational data and they directly integrate with the learning process. Machine learning algorithms are facing scalability and generalization across diverse graphs. GNN works on graph data structure, using unique IDs as nodes in a graph, with edges illustrating their relationships. Graph Neural Networks is used to improve the accuracy and efficiency of fraud detection by learning the complex relational information obtained from providers, beneficiaries, and physicians. We created a graph database based on the healthcare provider dataset. In this graph database, two types of heterogeneous nodes are there that are beneficiary and medicare provider nodes. The connection between the beneficiary and medicare providers is a power edge and the connection between providers is a shared-physician edge. We developed a fraud detection model using both machine learning and graph neural networks. Our Graph convolutional Network (GCN) model performed well compared to the basic machine learning (Logistic regression) model. The complex relationships between provider and beneficiary, provider and physician helped to detect medicare fraud using our model.