Machine learning model of tax arrears prediction based on knowledge graph

Most of the existing research on enterprise tax arrears prediction is based on the financial situation of enterprises. The influence of various relationships among enterprises on tax arrears is not considered. This paper integrates multivariate data to construct an enterprise knowledge graph. Then, the correlations between different enterprises and risk events are selected as the prediction variables from the knowledge graph. Finally, a tax arrears prediction machine learning model is constructed and implemented with better prediction power than earlier studies. The results show that the correlations between enterprises and tax arrears events through the same telephone number, the same E-mail address and the same legal person commonly exist. Based on these correlations, potential tax arrears can be effectively predicted by the machine learning model. A new method of tax arrears prediction is established, which provides new ideas and analysis frameworks for tax management practice.