Research And Application On The Medical Insurance Fraud

The reformation of national basic health care system makes health insurance become popular, also makes health insurance fraud become more serious. The social welfare has suffered huge losses, which has seriously affected people’s medical treatment. Therefore, it is particularly important to establish a perfect medical insurance fraud detection system quickly.The main work of this paper is to research from the medical insurance data, extract the appropriate features, build medical insurance fraud detection model based on the algorithm of data mining. Firstly, based on the knowledge of medicine and statistical analysis, this paper presents a feature extraction method. After cleaning the original medical insurance data, based on the professional medical knowledge and principle of statistics we put forward a method for constructing two level features, the features are common drugs and treatment rules; combined with the Gaussian mixture model and math integral method, we proposed a new deviation characteristic from drug categories. Then, according to the existing medical insurance fraud detection rules, the data set is divided into two aspects about normal and fraud data, and puts forward a method to cluster the normal behavior data, then make the normal data to some class clusters, so, we can build a classification model for each normal behavior cluster and fraud data. Based on the extraction of new features, this paper uses the Gauss mixture model to cluster analysis, after getting different cluster, we can use random forest and support vector machine classification algorithm to establish a classification model for each cluster and fraud data. After the establishment of the model, when comes with new data, the feature extraction is being first done, then we calculates the distance with clusters, and selected the nearest cluster classification model, so, we can use the classification model to fraud identification.This paper uses the data from Medical insurance administration, we use data to test our fraud detection model. From the experimental results we can know that our fraud detection model has a good effect on fraud detection. The results show that the proposed medical insurance fraud detection model has good performance in practical application.