Predicting Property Tax Classifications: An Empirical Study Using Multiple Machine Learning Algorithms on U.S. State-Level Data

This study presents a comprehensive analysis of property tax classification using machine learning approaches applied to the 2024 U.S. Property Tax Roll dataset. The research employs four different machine learning algorithms - XGBoost, Random Forest, Support Vector Machine (SVM), and Logistic Regression - to predict and analyze property classifications across American states. To address the challenge of imbalanced data distribution in property classes, we implement the SMOTE technique for data balancing. The experimental results demonstrate that the XGBoost algorithm achieves superior performance with an accuracy of 0.901, significantly outperforming other models across multiple evaluation metrics. The study reveals strong correlations between total assessment values and tax exemptions (correlation coefficient 0.98), providing insights into the relationship between property valuation and tax policy implementation. The findings have important implications for both tax administrators and policymakers, offering a data-driven approach to property tax classification and assessment.