One of the most valuable natural resources is water, which is essential for human development, economic growth, and environmental sustainability. In regions like Tangier-Tetouan, Morocco, rapid urbanization, population expansion, and climate variability have intensified pressure on water resources, making accurate water demand forecasting critical for sustainable management. This study reviews the evolution of water demand forecasting techniques, comparing and contrasting between (ML) Machine Learning and (DL) Deep Learning approaches with traditional statistical models, such as CNN-LSTM, Transformer-based models (e.g., PatchTST), and Gradient Boosting Machines (e.g., XGBoost). These models have demonstrated superior accuracy in capturing complex temporal and spatial patterns, particularly when integrated with external factors like rainfall, dam levels, and weather data.