Comparison of Logistic Regression and Random Forest Methods in Predicting Vehicle Tax Payment Compliance

Abstract

Motor vehicle tax is a major source of Regional Original Income (PAD). However, the level of motor vehicle tax payment compliance in North Aceh Regency is still suboptimal, particularly related to late payments. A data-driven approach is needed to predict and understand taxpayer compliance patterns more accurately. This study aims to compare the performance of the Logistic Regression and Random Forest methods in predicting motor vehicle tax payment compliance, as well as to identify factors that influence taxpayer compliance behavior at the North Aceh Samsat (Sat). This study uses secondary data in the form of motor vehicle tax payment transactions at the North Aceh Samsat for the 2022–2024 period, totaling 100,000 observations. The response variable is the tax payment compliance status (compliant and non-compliant), while the predictor variables include vehicle age, type of ownership, vehicle type, and vehicle brand. The data is divided into 70% training data and 30% testing data. The performance evaluation model is conducted using accuracy, precision, recall, and Area Under Curve (AUC) metrics. The analysis results show that Random Forest has better predictive performance than Logistic Regression, with higher accuracy and AUC values. Vehicle age and type of ownership are the most influential variables in predicting tax payment compliance, while vehicle brand has a relatively smaller influence. Logistic Regression provides a clear interpretation of the variable relationship, but has lower discrimination ability than Random Forest. Random Forest has proven to be more effective as a prediction model for motor vehicle tax payment compliance at the North Aceh Samsat. The application of machine learning-based predictive models has the potential to support more targeted policy making in an effort to improve motor vehicle tax payment compliance, especially in reducing late payments.