Research On Graph Neural Network In Medical Insurance Fraud

Medical fraud refers to the behavior of consciously obtaining illegal benefits in the field of medical insurance.Protecting against the risk of the provider fraud is critical for insurers for providers are more likely to take advantage of their duties to carry out fraud.The claim bill that the beneficiary must submit contains real medical scene information.There are many fraud detection algorithms,including supervised learning and unsupervised learning algorithms,which assume that each entity point in this scenario is independent of the other,which is inconsistent with the reality of social networks between people.The graph algorithm has obvious advantages in this scenario with mining the interaction information between different entities.However,in the real medical scene,too many entities will also lead to challenges in computing power.Therefore,this paper improves and innovates the graph neural network while preserving as much medical scene information as possible.Firstly,it is proved that the neighborhood matrix based on the meta-path in GTN can have the same prediction and high efficiency.Then a large number of isolated nodes in the graph are clustered by constructing the claims feature and graph embedding feature.By clustering labels and connecting edges,the similarity of suppliers in node groups can be more effectively measured.Finally,the model is trained to predict and explain the importance of different types of meta-path,which can help the insurer take targeted measures in risk control.The meta-path learning model based on the neighborhood graph can achieve an accuracy rate of 93.13% and F1 score of 66.47%,which is higher than the baseline model.The model achieves a higher recognition rate and guarantees the running speed,which provides the possibility to identify medical insurance fraud under the background of big data.