Journal of Tax Research

This study examines using data mining techniques to enhance tax evasion detection performance. Data mining is a process to discover uncertain, unknown, and hidden information from a database and it is a unique method of finding new facts and relationships in the existing data that have not been discovered by experts yet. In this study the usefulness of data mining based on association rules is employed as a tool to detect tax evasion. The research population consists of all listed firms on the Securities and Exchange of Tehran. The sample includes 125 firms which are selected by Cohesiveness Removal Method in the period of 1383 to 1390. In this study, 28 financial and non-financial variables in 9 classes have been used to build models. The association rules were used by apriori algorithm to detect tax evasion firms. For this purpose, the data were randomly divided into three categories: training, validation and test groups. The findings indicate that the method of data mining based on association rules by developing two models of 91% accuracy on training data, the percentage of 88% accuracy on data validation and the percentage of 86% accuracy on test data are able to detect tax evasion.