Research On Fraud Recognition Algorithms Of Basic Medial Insurance

In recent years,the universal medical insurance has been basically realized,and many works including real-time settlement are also in continuous progress.The following is that the monitoring situation of medical insurance service is facing more severe challenges,that is,the cases of fraud and deception of basic medical insurance(referred to as "medical insurance")have increased dramatically,which has brought huge losses to the national property,but also damaged the fairness and social Medical security.In order to strengthen the efficient and intelligent monitoring of outpatient,inpatient,drug purchase and other medical services,the national human resources and social security department has adopted many means,including artificial diagnosis and treatment rule screening,expert intervention and data ratio equivalence.These methods have achieved certain results,but due to the backward technical means,there are still great limitations.There is room for further improvement and upgrading in the application concepts,methods and technologies of big data.Therefore,whether we can combine the new feature extraction method and model fusion technology to effectively solve the identification problem of basic medical insurance fraud behavior(medical insurance fraud identification for short)has become the starting point of this study.Firstly,in view of the imperfection of the features extracted by the traditional methods in the basic medical insurance fraud recognition scene,this paper proposes a method of secondary feature extraction based on behavior similarity and multi category algorithm,which takes the intermediate results generated in the model training process as statistical features on the basis of traditional feature extraction,among which multi category algorithm includes polynomial combination and boosting tree model leaf.Sub node feature extraction and embedding feature extraction based on word2 vec.This method effectively avoids the problem of low quality of feature generation of traditional feature extraction methods,and achieves the effect of providing perfect and high quality features for the subsequent prediction model,so as to ensure the significant improvement of prediction performance.Secondly,aiming at the problem of unbalanced distribution of analogy in the scene of medical insurance fraud identification,this paper abandons the traditional technical means such as over sampling,down sampling and re scaling,and puts forward the method of sample division based on the idea of sample proportion equalization,making full use of the existing sample information,achieving the effect of sample proportion equalization,avoiding the problem of over fitting.Thirdly,a new model fusion algorithm(THBagging)is proposed to solve the problems of low recognition rate and weak automation of traditional medical fraud recognition methods based on the feature extraction method and new sample division strategy.This method has two-tier algorithm structure,using multiple groups of boosting based tree model algorithm for fusion,not only overcome the shortcomings of the existing algorithm structure unreasonable,but also effectively improve the intelligence of the identification method,and finally achieve the effect of improving the accuracy of basic medical fraud identification.Finally,the performance of the above methods is verified and compared on the real basic medical insurance data set,and the conclusion is given.