Development of a prediction model for tax assessments using data mining and machine learning tools.
Loading...
Date
2023
Authors
Sampa, Anthony Willa
Journal Title
Journal ISSN
Volume Title
Publisher
The University of Zambia
Abstract
A nation's economic development still depends heavily on its tax system. The tax administrations in Zambia much like many other countries face many difficulties in the tax collection process, with inadequate compliance being the main one. Therefore, in order to fix the issues and improve total collection, it is critical to be able to identify revenue leakages as much as feasible. In this study, we examined Zambia's tax audit and assessment procedure, which looks for income leaks brought on by fraudulent activity, understated assets, and declaration errors. It is crucial to carefully distinguish cases that are likely to result in smaller, insignificant revenue collection from those that yield significant revenue in comparison with the limited resources used to perform audits and assessments because of the large volumes of audit cases generated by audit selection methods, some of which yield very little collections after the audits. In order to identify the audit and evaluation choices that are most likely to result in a noticeably large revenue collection from these leakages, we created a machine-learning model employing supervised learning. Using the Random Forest technique, we created a prediction model that yielded a 91% ROC curve evaluation. The experiments' outcomes demonstrated that the prediction model created was able to accurately distinguish high value assessment cases for audit from the rest of the cases. This prediction model will help focus the auditing resources in areas that will yield the most revenue and ultimately assist the revenue authority in efficiently raising the nation's tax collections.
Keywords: Compliance, Audit, Assessment, Machine Learning, Tax
Description
Thesis of Masters of Science in Computer Science