Research On Medical Abuse Detection Based On Heterogeneous Network Community Division

With the advancement of medical informatization,medical insurance fraud cases have increased year by year.In the case of medical insurance fraud,the personnel who commit fraudulent acts are complicated and the means of fraud are numerous,which seriously endangers the safety of the medical insurance fund.China's medical insurance business started late,the number of insured people and the scale of medical insurance funds is huge,the problem of medical insurance fraud is more prominent,and the anti-fraud is strengthened to ensure the security of the fund is imperative.The problem of medical insurance fraud has always existed,but the existing methods do not solve this problem well.Rule-based detection methods require experts who are proficient in fraud detection research and familiar with medical insurance business are required to design rules.And fraudsters will try to avoid detection rules as much as possible.Anomaly detection is one of the main techniques for fraudster detection.This type of method focuses on finding global outliers and cannot effectively detect local outliers.In the detection of medical fraudsters,it is difficult to effectively cluster them according to the statistical characteristics of doctors and patients.Therefore,the existing detection methods do not solve the problem of fraud detection well.Medical insurance data is intricate and unevenly distributed.Most medical records are not complete for the personal information records of doctors and patients.Traditional methods are difficult to effectively cluster according to the statistical characteristics of doctors or patients.As a clustering method,the network community partitioning method is used to discover the community structure in the network,transform each individual of the problem space into nodes on the network,and pay more attention to the relationship between the nodes rather than the attributes of nodes in the clustering process.This method is very suitable for solving the problem of medical insurance fraud detection.At present,this method is mostly applied to Homogeneous information networks.Since there are many types of entities in the field of medical insurance,consider using heterogeneous network community partitioning methods to solve the problem of medical insurance fraud detection.This paper has made some effective research on the detection of abnormal subjects in the field of medical insurance.The main work and contributions are as follows:1.It is proposed to model medical insurance data as a heterogeneous weighted network.The medical insurance data includes doctors,patients,medicines and other subjects,abstracting doctors,patients and drugs into nodes in a heterogeneous network,and assigning weights according to their relationship.Putting data into a heterogeneous network can more clearly find the relationship between the subjects,which is conducive to the resolution of fraud detection problems.2.A fraudulent patient detection method based on label propagation algorithm FDBLPA is proposed.The medical insurance data was constructed into a patient-drug heterogeneous weighted network;the network was divided using the improved overlapping community label propagation algorithm NSLPA;finally,abnormal patients were detected in the divided communities.The method initializes the node according to the disease code,which reduces the randomness of the original method.At the same time,it designed the label propagation rules that conform to the data characteristics,which improved the accuracy of abnormal patient detection.3.A fraudulent doctor detection method based on modularity optimization algorithm FDMOA is proposed.For general hospitalization data,doctors and drugs are modeled as heterogeneous networks;the modularity optimization algorithm FNO was used to classify them into corresponding communities.Finally,an abnormal doctor was discovered through a comparison between the doctor community and the drug community.According to the doctor's department and the disease code,the community is initialized and divided into doctors and drug nodes,which improves the efficiency of the algorithm.At the same time,in the process of node merging,different edge weights are considered,which improves the accuracy of the algorithm.