Analysis Of Medical Abuse Fraud Detection Based On Graph Mining

In recent years,Healthy China has gradually risen to be a national strategy,and health insurance construction occupies an important position in economic and social development.With the popularization and promotion of medical information,healthcare fraud has been recognized as a serious social problem.Medical abuse is a major fraud in healthcare fraud,which mainly refers to medical institutions or doctors to provide medicines or medical supplies and the actual treatment with inconsistent or contrary to medical standards,so as to increase spending on health care.A variety of healthcare insurance fraud cases,which greatly damaged the interests of the insured,constitute a major threat to the safety of medical insurance policy in various countries.Although,fraud is not a phenomenon happened recently,fraud detection problem still has not been solved very well.Firstly,some traditional fraud detection approaches are based on detection rules to find out violations of these rules.However,the performances of these traditional approaches are inherently limited by the domain knowledge of experts.Furthermore,owing to the skewed class distribution[6]of healthcare data,extremely little fraud records make it difficult to identify fraudulent records from a large number of normal records.Next,healthcare datasets have been evolving dynamically with various inner and outer spatial fact over time and new fraud patterns will emerge constantly to circumvent the fraud detection found.Lastly,although there are many literatures have proposed kinds of different approaches to solve fraudulent problems in various individual areas,these approaches do not always work well.The supervised approaches in these literatures concentrate on defining the fraudulent problem as a binary classification problem.In order to generate a more accurate classification result between fraudulent records and normal records,volume profiles of the entities involved in the data should be analyzed.This work costs a lot of effort and may violate privacy policy in healthcare domain.While unsupervised methods such as outlier detection and clustering may not be accurate as supervised methods because they do not take additional information into account.To sum up,a new healthcare fraud detection method is needed which involves and has a higher degree of accurate.Our main works and contributions are as follows:1.This thesis proposes a healthcare fraud detection method based on trustworthiness of doctors,GM-FP.By the trustworthiness of doctors,this paper combines graph mining with frequent pattern mining algorithms together,which only uses healthcare records to train a reasonable treatment of a disease model(type,quantity and the relationship between drugs and medical facilities).Based on the trustworthiness and frequent pattern of each doctor from the healthcare records,this paper constructs a rational treatment model to distinguish anomaly from normal records by calculating the similarity between the model and the unknown record.2.This thesis proposes a healthcare fraud detection based on intrinsic feature and network exploration,IF-NE.For each healthcare record,IF-NE chooses appropriate classifier to classify the normal and abnormal records using the internal features and the features based on the characteristics of the network,and then determines the healthcare record is a record of fraud.The internal features are obtained based on RMF(new schedule,frequency and amount spent).And feature extraction based Network has enriched the doctor-patient bipartite graph and build doctor-patient-records tripartite graph with healthcare records.Furthermore,IF-NE use a influence propagation algorithm to infer the scores of network components(i.e.the patient and the doctor health records)which start the algorithm from a mark edges(i.e.health records fraud)as network characteristics.Finally,the results show that the method is more effective than the benchmark method using the random forest based on the features extracted from IF-NE.3.This thesis proposes a healthcare fraud detection method based on non-convex label propagation,NCLP,in healthcare data set.This method improved label propagation method based on the traditional convex label propagation.And through the convex concave convex transform,NCLP transforms label propagation algorithm to non-convex label propagation algorithm for healthcare fraud detection,which can solve the problem of performance reduction when using convex label propagation class in imbalanced healthcare data set.