Comparison of Decision Tree (C5.0 Algorithm and Random Forest) and Support Vector Machine in the Validation of Taxpayers

Today, the risk-based audit method is emphasized in modern tax systems, so explaining a comprehensive model for rating the risk of taxpayers is one of the basic steps of implementing a comprehensive tax plan. Therefore, in this article, we aim to measure the performance of decision tree algorithms and support vector machine in the validation of taxpayers. The statistical population of this research is the companies accepted in the Tehran Stock Exchange, which were active during the years 2012-2017 and for the selection of the sample was made using the screening method (elimination). In this research, first, using Delphi technique and meta synthesis, 164 effective components in the validation of taxpayers were identified, then the data needed to measure the variables of the research were extracted from the Kodal website and by examining tax files, and finally by using the collected data, we investigated the accuracy of the decision tree (C5.0 algorithm and random forest) and support vector machine in validating taxpayers. The findings showed that based on the results of the AUC value, the C5.0 algorithm and the random forest have a better fit, however, the research hypothesis that it is possible to predict the risk of taxpayers using the SVM algorithm is not rejected.