Application Research Of Machine Learning Algorithm In Tax Risk Monitoring Platform

Tax risk monitoring is a key and difficult task in the development of China's taxation industry.The rapid development of the Internet and information technology has brought great convenience to tax risk monitoring.A good tax risk monitoring mechanism can not only reduce the complicated and complicated risk investigation work of tax officials,but also improve the response of tax personnel in risk response.Quality effect.In the risk evaluation work of the risk evaluation case sampling evaluation(hereinafter referred to as case evaluation),in the face of a large number of risk cases that have been completed,there are two main problems in the existing case evaluation working mechanism:(1)the sampling method cannot As much as possible to extract risky cases,resulting in leakage of tax issues;(2)in the face of tax risk monitoring platform(G tax bureau)risk identification is accurate,tax personnel should be consistent with the platform identification results,the final case evaluation The result is a problem of repeated responses to satisfactory cases.The traditional sampling method can not effectively improve the hit rate of risk cases,and the case evaluation process is complex and the workload of personnel is large.This paper combines the related problems generated in the case evaluation to study the use of machine learning technology in the tax risk monitoring platform.The construction of indicator system and the generation of feature data are introduced.The evaluation results and risk characteristics in case evaluation are used as sample data.The sample data is divided into two parts: “satisfaction” and “unsatisfactory”.The sample data is modeled by a related classification machine learning algorithm,and the feature variables are filtered according to the IV(Information Value)statistic to improve the prediction accuracy of the model.Finally,the predictive power of each classified machine learning model is evaluated using an evaluation index such as AUC.Through the analysis of the experimental results,the risk characteristic data generated by the indicator system,the logistic regression algorithm after GBDT feature processing is suitable for the modeling of case evaluation data,which can effectively distinguish the satisfaction of the risk response case in the evaluation process.