Accurate forecasting of household water demand is essential for sustainable urban water management, particularly in resource-constrained cities like Rajshahi, Bangladesh. This study addresses a key research gap by developing a machine learning–based prediction model using real-world data from 15 households equipped with overhead tanks. Daily water consumption data were collected over six months, incorporating environmental and socio-economic variables such as household size, income, temperature, rainfall, and weekly patterns. After preprocessing and feature engineering, two models, Linear Regression and Random Forest, were evaluated. The Random Forest model outperformed Linear Regression with a lower Mean Absolute Error (9.27 L), Root Mean Squared Error (11.70 L), and a higher R² score (0.9951). Feature importance analysis identified household size, income, and temperature as primary predictors. Seasonal and weekly patterns were evident, with higher demand during summer months and weekends. The model's high accuracy and low feature requirements make it scalable and practical for urban utilities. Existing literature shows that proper water supply scheduling can reduce wastage by an estimated 10-15%. The study aligns with its objective of enhancing predictive capability using data-driven methods and offers valuable implications for water resource planning in rapidly urbanizing areas.