Intelligent Classification of Residential Property Tax Levels Based on Temporal Clustering and Federated Learning

Abstract

To address the problem of inconsistent and highly sensitive property tax level classifications across multiple U.S. states, this study proposes a distributed intelligent classification model based on temporal feature clustering and federated learning. The method first uses time windows to analyze property transaction histories and tax records and constructs typical property patterns. Then, a federated framework is used to jointly train classifiers across multiple state governments, avoiding data leakage. The model achieves an average accuracy of 87.5% on validation sets from seven states and can automatically adjust tax level ranges based on land price trends.