Predicting Household Piped Water Consumption Using Satellite and Street View Images in Two Indian Cities

Dataset supporting machine learning models predicting household water consumption and income from satellite imagery, Google Street View semantic segmentation, and geospatial features in Hubballi-Dharwad, India (n=21,702 households for income; n=15,273 for consumption). Includes processed features (GSV segmentation, image embeddings, nightlight, population, building size), ordinal outcome variables (water consumption and income categories), trained models, and spatial identifiers. Enables replication of LightGBM models achieving 73% accuracy (ROC-AUC 0.79-0.89). Data collected January-May 2023 as part of Karnataka Urban Water Supply Modernization Project (KUWSMP) Impact Evaluation. Raw imagery excluded for privacy; processed features and code provided for model replication and transfer learning studies.