Analysis of Taxpayers with a Data Mining Approach

Abstract

Analysis of taxpayers' behavior is important in maintaining tax justice and strengthening the foundations of the tax system. By better understanding the behavior of taxpayers, appropriate measures can be taken to prevent tax evasion. In this study, tax payers were analyzed with data mining approach. In this regard, X-Means and K-Means algorithms were used to cluster taxpayers through RapidMiner software and the information of 9994 taxpayers in several business categories. Based on the results, the number of optimal clusters was seven in the Kamiangin method, and the average number of optimal clusters was three in the X method. Examining the clusters in the Kamyangin method shows that in cluster (7), the amount of tax contribution expressed is lower than other clusters. In this cluster, there were businesses related to laboratories, radiology, physiotherapy, etc. On the other hand, the highest share of tax expressed from diagnosis also belonged to cluster (1). In this cluster of money changers; Guild of cloth bankers; bags and shoes, bags and suitcases; notary offices; audio and video equipment; There were sweets, nuts and ice cream. Based on the X-mean results, taxpayers were classified into three clusters, and the largest share of the studied indicators in cluster one includes money changers; Guild of cloth bankers; bags and shoes, bags and suitcases; notary offices; audio and video equipment; Sweets, nuts and ice cream; vehicles and spare parts; Dentists. Also, the decision tree classification method was used to predict the final tax. Based on the results in the prediction of the final tax, the highest weight is assigned to the variable of zero declaration ratio, declared tax share and finally the number of taxpayers. Also, the accuracy of the artificial neural network has been obtained at 97%, which shows that it is a more characteristic approach to clustering with an accuracy of 66.67%.