Implementation of Random Forest Algorithm for Classifying Land and Building Tax Arrears and Risk Factor Analysis Dashboard

Abstract

This study aims to develop a predictive model to identify the potential for land and building tax arrears and analyze the dominant risk factors contributing to non-compliance. The research utilizes the Random Forest classification algorithm applied to historical tax data from the Regional Financial and Revenue Management Agency of Binjai City. The approach involves data preprocessing, feature engineering including target encoding for geographical areas, and model training with hyperparameter tuning to optimize classification performance. Furthermore, a web-based interactive dashboard is developed using the Flask framework to visualize the predictions and risk factors. The results demonstrate that the Random Forest model achieves a robust and consistent accuracy of approximately 85% in classifying compliant and non-compliant taxpayers. Feature importance analysis reveals that land area is the most dominant risk factor influencing tax arrears, significantly outweighing other variables. In conclusion, the integration of the Random Forest algorithm with an interactive dashboard provides a highly accurate, efficient, and scalable solution for local governments to transition from reactive tax collection to proactive, data-driven risk management.