Assessing Taxpayers' Ability to Pay

A Machine Learning Approach

Abstract

Tax revenue remains one of the challenging fiscal issues in Indonesia. Improving tax collection performance through comprehensive reform has been an influential agenda, especially for the Directorate General of Taxes. One of the critical improvement areas is the utilization of information technology in tax assessment and audit functions. This study explores the taxpayers’ ability concept as a complementary measure to the existing taxpayer monitoring module, particularly in case selection and targeting functions under the Compliance Risk Management (CRM) framework. The 5Cs of credit analysis (Character, Capacity, Capital, Condition, and Collateral) are employed as proxies for the taxpayers’ ability to pay. This research aims to identify the most effective machine learning algorithm for classifying taxpayers' ability to pay to enhance the CRM's effectiveness for corporate taxpayers, limited to those administered in large and medium tax offices. Several machine learning algorithms were tested, including logistic regression as a baseline comparison, based on the quantitative and qualitative performance comparison. The findings reveal that the Light Gradient Boosting Machine algorithm provides the most effective results in terms of both accuracy and computational efficiency. However, several challenges need to be addressed to improve the model implementation.