Purpose This paper aims to address the key limitations in urban water demand forecasting models, such as overfitting, limited interpretability and instability when confronted with nonlinear and nonstationary patterns. By developing a hybrid framework that integrates the Consumption-Based Sustainability Index (CBSI) with advanced machine learning (ML) methods, the study seeks to deliver robust, accurate and generalizable daily water demand predictions for municipal pumping stations. These reliable forecasts bolster operational decision-making, optimise resource allocation and enable proactive management strategies, thereby improving the performance and resilience of water distribution systems in rapidly growing cities. Design/Methodology/Approach A hybrid framework was developed, combining CBSI-guided feature selection with ML models – Random Tree (RT), Polynomial Kernal (PUK), KSTAR, Random Forest (RF) and Locally Weighted Learning (LWL) – across six major pumping stations in Pune. The RT model consistently provided the highest accuracy and resilience across varying demand patterns. This scalable, data-driven approach supports utilities in optimizing pump schedules and managing shortages, while also offering adaptability for broader water resource management. Findings Among tree-based (RT, RF), instance-based (KSTAR, LWL) and SVM-based (PUK) models, the RT model achieved superior performance across all stations, effectively forecasting complex, nonlinear, and nonstationary demand patterns. Its success is attributed to recursive partitioning, local adaptability and CBSI-driven hyperparameter and feature selection process. Original Value By overcoming longstanding modeling limitations, this hybrid CBSI–ML framework advances the field of urban water demand forecasting, thereby supporting aiding sustainable, resilient, and efficient water management system rapidly urbanizing environments.
Leeladhar Shankar Pammar (Fri,) studied this question.