大数据背景下税收风险识别精准度存量研究——基于机器学习的视角  

The Stock of Research on Accurate Identification of Tax Risk under the Background of Big Data Technology——Based on Machine Learning

作　　者：姬颜丽[1] 王文清[1] Ji Yanli;Wang Wenqing

机构地区：[1]国家税务总局税务干部学院

出　　处：《财政研究》2020年第9期119-128,F0003,共11页Public Finance Research

基　　金：国家税务总局税务干部学院青年课题“大数据技术背景下完善税收风险管理体系建设的研究”(项目编号:2020QNKT005)的阶段性研究成果。

摘　　要：税收风险识别对税收风险管理至关重要。无论是业务驱动的传统税收风险识别还是数据驱动的分析模型,都很难从单方面来提高税收风险识别的精准度。利用新技术提高税收风险识别的精准度成为当务之急,机器学习成为可选项。本文认为,利用风险应对核查成果的效力,把机器学习运用到税务机关防范税收风险的识别中,可实现人工经验和大数据分析的双轮驱动,为提升税收智能化管理开辟新的视野路径。本文选取随机森林算法,以商贸企业为例,对虚开增值税发票风险建立识别模型,通过检验推演验证得知,该模型稳健可靠、预测准确性高,可供税务机关参考借鉴。Tax risk identification is important for tax risk management.Neither the traditional business-driven tax risk identification nor the data-driven analysis model can improve the accuracy of tax risk identification unilaterally.Therefore,we put forward that we must make full use of the results of risk response verification,apply machine learning to the identification of tax authorities to guard against tax risks,realize the two-wheel drive of manual experience and big data analysis,and open up a new visual path for improving the intelligent management of tax revenue.In this paper,the random forest algorithm is selected to establish the identification model for the risk of false invoicing of business enterprises.Through the test,the model is robust and reliable,and the prediction accuracy is high,which can be widely used.

关 键 词：大数据 机器学习 税收风险识别 随机森林 虚开发票 

分 类 号：F812.45[经济管理—财政学]