The application of decision tree and artificial neural network to income tax audit: the examples of profit-seeking enterprise income tax and individual income tax in Taiwan

Abstract

Income tax is one of the key revenues for a government, though taxpayers are often tempted to evade it, using many different means. In Taiwan, income tax auditing relies on a computerized selection of cases. Its result is thereby dependent on the quality of sampling, so is its effect on curbing tax evasion. Insufficient workforce also makes the auditing more difficult. To enhance the effectiveness of tax auditing, we used previous filing records of enterprise and individual income taxes as a database to develop two different models via decision tree and artificial neural network. The accuracy rates of the models were compared in order to confirm their suitability for different tax types. The results showed that the DT model had a higher accuracy rate in detecting tax evasion in individual income tax. In comparison, ANN performed better in the category of enterprise income tax. Auditors have to choose different techniques for different tax types, either based on their own experience or additional information, such as the results of this research, which provides a useful reference for the choice of auditing tools.