An Enhanced Framework for Urban Water Consumption Analysis: Feature Clustering with Ensemble Methods

Urban water consumption analysis presents significant challenges due to the complex interplay of socioeconomic, demographic, and built environment factors. This paper introduces a novel Feature Clustering Framework of TopK and Threshold with Ensemble Method (FCTTE) specifically designed to address high-dimensional urban datasets. We evaluate this framework using a comprehensive dataset of 1,120 features across eight domains related to New York City's urban environment. Our experiments demonstrate that FCTTE significantly outperforms conventional feature selection methods, improving LightGBM classification accuracy by 4.6% compared to baseline, while traditional methods achieved only 1% improvement. The framework identified median family income, energy usage intensity, adult male population, greenhouse gas emissions, and commercial building characteristics as the most influential factors affecting water consumption. By effectively managing feature redundancy through hierarchical clustering and strategic selection, FCTTE provides urban planners with interpretable insights for water resource management while maintaining superior predictive performance. This integrated approach bridges the gap between fragmented analyses of individual urban factors and the need for holistic understanding of water consumption patterns in complex urban environments.