Predictive analytics and its impact on tax audit case selection in revenue authorities of developing economies

Abstract

Revenue mobilization, especially in developing economies, is an arduous task for tax authorities. Uganda Revenue Authority (URA) uses manual, intuitive methods for Corporate Tax audit case selection, which could be time-consuming and ineffective. The study aims to gain insights into the efficiency of predictive analytics in the selection of tax audit cases based on historical tax audit data and taxpayer attributes. Tax officials of the Uganda Revenue Authority (URA) were interviewed, and data analyzed using machine learning algorithms after being trained on 5 years' worth of audit data (2017–2021) provided by URA. The objective was to evaluate and validate the application of predictive analytics for enhanced audit case selection at URA in order to better identify high-risk taxpayers, improve resource allocation and address the operational inefficiencies associated with current audit selection methods. Three models were tested: an initial leaked model (~99% accuracy, AUC 0.99, but unrealistic), a baseline logistic regression model (~62% accuracy, AUC ~0.68), and a final tuned XGBoost model (~75% accuracy, AUC ~0.80, precision ~75%, recall ~85%). The XGBoost model demonstrated substantial improvement over the baseline, correctly flagging a higher proportion of high-risk entities with practical accuracy and balanced performance. The results indicate that high-risk entities selected by the predictive model exhibit a significantly higher likelihood of non-compliance compared to those selected through traditional methods. This demonstrates the potential of predictive analytics to increase audit efficiency and yield while optimizing scarce enforcement resources. This paper joins a recent stream of literature on the application of data mining in tax administration. It contributes to the understanding of how predictive analytics can enhance tax audit efficiency in developing economies, and specifically provides empirical evidence and guidance for policymakers in Uganda and other similar contexts considering data-driven audit selection frameworks. Keywords: Predictive Analytics, Tax Audit, Audit Efficiency, Revenue Mobilization, Uganda Revenue Authority