Subject/s of Focus: Machine Learning

Authors: Gilbert Chua, Chelsea Ong, Jomilynn Rebanal, Van Yu

This project is for the fulfillment of the requirements in our Machine Learning Course.

Abstract

Tax, the lifeblood of government, enables the provision public goods and services to its citizens. Having an adequate amount of taxes is essential for a government to finance different initiatives like infrastructure development and social welfare support.

This study intends to predict if a large taxpayer (LT) will file their annual return using their filing behavior in the preceding quarters combined with their location. These features were chosen under the assumption that recent behavior can reveal a taxpayer’s predilection to non-compliance.

Using the Gradient Boosting Classifier and Balanced Accuracy Score (BAC) as the scoring metric, we have obtained a 97.10% accuracy in predicting taxpayer non-compliance. The best predictors were the amount paid in the 2nd quarter if filed, amount paid in the 3rd quarter if filed, and filing or non-filing the 3rd Quarter. This model yielded a recall score of 81.71% and a precision score of 87.21% when classifying the taxpayers who will not file their annual return.

Based on these findings, we highly recommend the development of an automated reminder system to bolster higher compliance in the annual return filing and payment. Moreover, this machine learning model can be integrated in the monitoring system to guide the implementation of more strategic and cost-effective interventions.

Note:

For more information regarding this project, you may send me a message via Linked in