Development of water consumption prediction models: a data-driven machine learning and deep learning approach

Water demand is increasing day by day due to rapid growth in urban areas and varying climatic conditions. Therefore, it becomes necessary to manage water resources in a sustainable way. Water consumption prediction can greatly support sustainable water resource management. In this study, our main objective is to identify more accurate and reliable techniques for water consumption prediction using machine learning (ML) and deep learning (DL) algorithms. We have evaluated various ML models, including K-Nearest Neighbours (KNN), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Adaptive Boosting (AdaBoost), as well as DL models like Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Deep Neural Networks (DNN), for predicting water consumption based on various input parameters. A hybrid model approach (XGBoost-LSTM) is also tested and evaluated based on its tree-based feature extraction and sequential pattern learning features. The dataset is divided into two separate sets for training and testing purposes to ensure the reliability of the predictive performance of the system and enable true out-of-sample evaluation. The accuracy of the models is calculated by mean absolute error (MAE), mean squared error (MSE), and the coefficient of determination (R2). The DNN model performs best across all models, with the lowest MSE of 1.35 and the highest R2 value of 0.99994, and it demonstrates a strong capability to capture complex nonlinear relationships in water consumption data. The proposed hybrid XGBoost‑LSTM model also demonstrates high prediction accuracy with its feature interactions and temporal dependencies. The outcome of this study reveals that both the DNN and hybrid model approaches show enormous potential for improving the performance of water consumption prediction. The proposed study does not include cross-regional generalisability, real-time data streaming and analysis, or additional climatic variable analysis that can be further explored in the future.