An Empirical Method for Discovering Tax Fraudsters

Abstract

This work encompasses the development of a new method for classifying tax fraudsters based on fraud indicators. This work was developed in conjunction with a Brazilian fiscal agency aim at avoiding fiscal evasion. The main contribution of this paper is a method that allows classifying and ranking taxpayers analyzing fraud indicators obtained from several fiscal applications. Particularly, we developed a method for identifying frequent fraud patterns using association rules and then we apply two dimension reduction methods (i.e. PCA and SVD) in order to create a fraud scale, which allows ranking taxpayers according to their potential to commit a fraud. Experiments were conducted using real taxpayer data. Tax auditors, specialized in fraud detection, validated our results. Preliminary results show that our method may indicate fraudsters with 80% of accuracy, which is definitely an excellent result.