Hyperparameter: optimized XGBoost for water demand forecasting using temporal and contextual features

Water demand forecasting based on utilization patterns and scarcity indicators requires careful consideration of temporally fluctuating factors. These factors including festivals, maintenance schedules, weather variations, and public holidays significantly affect water demand patterns. To address this challenge, this article proposes a hyperparameter-optimized Extreme Gradient Boosting (XGBoost) model for water distribution and demand forecasting. The proposed model employs systematic hyperparameter tuning to enhance predictive performance and mitigate overfitting. Tuning targets three core functions: feature engineering, feature scaling and regression-based demand prediction. These functions are optimized by selecting precise hyperparameters that improve predictive performance. The model is evaluated using historical water consumption data collected from the Lawspet region, Puducherry, India (2017–2021), and compared against ARIMA, SARIMA, and several state-of-the-art machine learning–based forecasting methods. Experimental results show that the proposed model achieves the highest R2 score among all benchmark models, reduces RMSE by 11.29%, and lowers runtime by 36.55% relative to the strongest baseline, evaluated on the full 60-month dataset.