Deterioration of water quality in surface and groundwater systems poses increasing challenges for public health and water resource management. Although water-quality indices and classical statistical methods remain widely used, they often fail to capture the nonlinear interactions and spatiotemporal variability inherent in complex aquatic environments. This study systematically reviews and synthesises recent applications of machine learning and deep learning models for water-quality prediction, classification, and regulatory assessment. A structured literature search process, coupled with transparent inclusion criteria, was used to identify studies spanning conventional regression and classification models, spatiotemporal DL architectures, and hybrid CNN–LSTM frameworks. The results show that ML models, particularly Random Forest, Support Vector Machines, and Gradient Boosting, consistently outperform traditional methods across a range of physicochemical parameters. In contrast, DL and hybrid models offer superior capability in capturing spatial and temporal dependencies. However, the review also reveals key methodological gaps, including limited cross-regional generalisation, inconsistent preprocessing steps, inadequate evaluation of WHO threshold exceedances, and insufficient attention to real-world challenges such as sensor drift and missing data. These findings underscore the need for improved benchmark datasets, more transparent modelling workflows, and broader adoption of explainable AI tools to enhance interpretability and regulatory trust. The study provides a comprehensive assessment of the current landscape of ML/DL water-quality modelling and outlines pathways toward more reliable, generalizable, and field-ready intelligent monitoring systems.
malisaba et al. (2026) studied this question.