The Detection of Taxpayers with False Invoices using Data Mining Techniques

In this paper, we present indices by which it is possible to characterize and detect those potential users of false invoices in a given year, depending on the information of their tax payment, their historical performance and characteristics, using different types of data mining techniques. In this research first, clustering algorithms like Self- Organizing Map (SOM) and neural gas networks are used to identify groups of similar behaviors of taxpayers. Then decision trees, neural networks and Bayesian networks are used to identify those variables that are related to conduct of fraud and/or no fraud, detect patterns of associated behavior and establishing to what extent cases of fraud and/or no fraud can be detected with the available information. We utilize some information gained from tax auditors who are working in the Tax offices of Tehran and the informal unofficial statistics and anonymous questionnaire from some companies to gain primary data to detect fraud and compare different techniques of false invoices. To determine the main indexes in false invoices, we divided taxpayers to the micro and small enterprises and on the other side medium and large enterprises and examined the factors of fraud on each groups, with neural gas networks, separately. Particularly the neural gas method found that it was possible to identify some relevant variables to differentiate between good or bad behavior, not necessarily associated with the use and sale of false invoices. Kohonen’s method however, did not provide any behavioral patterns. In the case of micro and small businesses, the percentage of correctly detected fraud cases was 92%, while in the case of medium and large enterprises, this percentage was 89%.