Data Mining for Detection of Tax Evasion: The Case of Tax Payers in Addis Ababa

Abstract

The Tax has a high contribution to an economy; the government uses tax revenue for different government expenditure. Businesses and privates have obligations to pay tax from their income to the government. Despite this importance and responsibilities, corporates and individuals are involved in tax evasion. In Ethiopia Specifically in Addis Ababa, this problem is severe that about 50% of companies are involved in tax evasion. This study is conducted to develop tax evasion detecting techniques by using data mining procedures. It has used data about taxpayers in Addis Ababa and collected from the ministry of revenue at different tax payer’s branch offices in Addis Ababa. The study has followed the KDD method of data mining. The study has conducted two main procedures for model development; cluster modeling and classification modeling. The cluster modeling was conducted by using the K-mean algorithm and classification modeling was conducted by implementing different classifiers; J48, Naïvebayes, Neural Network, and Random Forest. Finally, the tax evasion detecting model was developed by using the Random Forest algorithm after making the comparison with other classifiers implemented. Besides, the decision rule construction was conducted by using the J48 algorithm. Finally, the study indicated that tax evasion practices with related to the liability of companies, expense, and amount of tax.