A Predictive Analytics for Detection of VAT Taxpayers Evasion Through Classification & Clustering Algorithms

Tax evasion is one of the ongoing challenges for any tax system, particularly for developing countries. The purpose of collection of value added taxes, based on its inherent characteristics of self control and capacity of tracking the transactions chain, is the gradual clarification of economical transactions, and also built-up of a new source of income that is sustainable and reliable for financing the government’s budget. Furthermore, it is necessary that such clarification or transactions among the traders be dominant since the first process of purchasing the raw materials up to the production and sale of goods, so as the tax may be rightfully collected.
The business Intelligence in general, and data mining in particular, are effective tools for enhancing effectiveness and efficiency in the assessment of tax evasion. In the present research, based on the available information on VAT taxpayers returns in the years of study (2009-2014), being audited by the tax administration, and data mining methods including, classification algorithms of Decision Tree, Naive Bayes, and K-Nearest Neighbor, and clustering algorithms of K-means and K-medoids, the tax evasions were predicted; then by applying the silhouette index the validation of results were performed, and in view of analysis on the tax indices of earned clusters, they were divided into low-risk and high-risk taxpayers. The results arising from applying the classification and clustering algorithms may assist the tax administration in planning for the assessment and prevention of tax evasion.