Research On Identification Of Social Medical Insurance Fraud Based On Random Forest And GBDT

Social health insurance is a compulsory social insurance system enforced by national laws and regulations to provide basic medical care to workers covered by social medical security.At present,the social health insurance in China mainly includes basic medical insurance and large-scale medical assistance,enterprise supplementary medical insurance,and individual supplementary medical insurance.With the enhancement of China's comprehensive national strength,the state's investment in social health insurance has been rising year by year.Meanwhile,the fraud in the process of social medical insurance reimbursement is increasingly frequent and the total amount involved is also growing,which make the social medical insurance fraud an urgent issue.Based on the current principles of data generation in medical diagnosis and treatment,the policies and regulations of health insurance,and the optimization of scientific methods,this study starts with a series of reasonable data cleaning and comprehensive statistical description so as to preliminarily determine the severity of social medical insurance fraud and the common characteristics of violators.Then,combined with business knowledge and statistical methods,a feature system that can fully reflect the information carried by the data is constructed from the two perspectives of global features and local features.With the employment of the proposed feature system,a composite model based on two integrated algorithms,random forest and GBDT,is built to solve the problem of intelligent auditing in social medical insurance fraud.According to the test results,the composite model proves to be effective in the improvement of audit efficiency and accuracy,which has achieved the expected effect.The main innovations of this paper are as follows:on the one hand,in the stage of feature engineering,the combination of global features and local features is used to show the data features more comprehensively,while retaining the information of the original data more completely;on the other hand,based on the two algorithms of random forest and GBDT.A compound algorithm model is constructed to further improve the prediction accuracy of the algorithm.