Research On The Risk Prediction And Optimization Of Enterprise Tax Arrears

Taxation is the main source of a country's fiscal revenue and an important indicator of a country's economic capacity.Tax collection is a work that all countries in the world value very much,it relates to the amount of national fiscal revenue.However,enterprises in the production and operation process,will not always be profitable for economic benefits.Some enterprises will pay less tax or not to pay taxes,these are tax arrears.The act of tax arrears will lead to the reduction of state revenue,including the disorder of national economic order,and the harm to the interests of the country and the people.Before the tax authorities make tax assessment of enterprises,they can not accurately identify whether the enterprises owe taxes.The tax authorities can not make tax assessments on all enterprises,so it is of great significance to predict the risk of enterprises' tax arrears and screen out suspected tax-owed enterprises to protect the state's fiscal revenue.At present,there are two main parts to the tax source enterprise's tax risk prediction:First,the tax authorities determine the enterprise data in the big data intelligence platform to identify the risk points.Second,tax experts judge by experience and knowledge of the risk points identified by the early automation manual judgment.That is to say,the current tax risk forecast automation part of the recall rate is high,the accuracy is low and still rely on the manual 'experience of tax experts.With the increasing number of tax source enterprises,and the tax management information system on the line,more and more relevant data,only simple rules and manual experience knowledge is difficult to complete the task of risk prediction.This paper studies the problem of high but low accuracy of the current enterprise tax risk prediction and recall rate,relying on manual experience determination.We will use the enterprise tax and social security data and data mining(Data Mining)method to predict the enterprise tax risk.We will try to eliminate the dependence on the labor experience determination to improve the efficiency of the tax risk prediction.The main tasks of this article include:1.Using corporate tax and social security data,we use the various data mining classification models built into the Weka tool to predict the risk of corporate tax arrears.Mainly includes:extracting some enterprise tax and social security data from the big data intelligence platform of tax authorities,pre-processing it,modeling and testing the processed data using the classification algorithm built into weka software,selecting F1 score as the evaluation index of the risk prediction problem of tax arrears,and comparing the prediction effect of each algorithm.In our experimental,we found that,the random forest algorithm has the best effect on the prediction of enterprise tax risk.2.In the context of big tax data,a large amount of corporate tax and social security data will lead to model training time is unacceptable.In this paper,the random forest algorithm is optimized in parallel for the disadvantages of the time-consuming training of random forest models.Based on a full understanding of the principle and structure of the random forest algorithm,two ideas of parallel optimization of the random forest algorithm are put forward:parallel between the decision tree and the decision tree internal parallel,and the two ideas are combined and applied.The above parallel optimization scheme is realized by Java code,and the enterprise tax and social security data are also used in the 8 CPU machines,and the experimental results show that the parallel optimization scheme can make use of multi-core resources to speed up the training process of random forest model to some extent.3.Aiming at the cumbersome shortcomings of manual use of Weka GUI,I will implement a data classification and automatic processing system for Weka API and integrate the improved random forest model into the Weka software to solve the problem of manual continuous operation when using Weka software.By calling the API provided by Weka software,the optimized random forest model is indirectly called using program to automate model training and prediction.Users can automatically complete the model training and prediction tasks with a single click,and reduce the steps of manual operation of Weka software by providing the training data set and candidate enterprise test set of enterprise tax and social security.