Machine learning (ML) and Internet of Things (IoT) technologies are rapidly reshaping how water quality is monitored and managed. This review synthesizes recent advances in IoT–ML applications across a range of aquatic environments, including rivers, lakes, groundwater, drinking water networks, wastewater treatment plants, bathing waters, aquaculture, and coastal systems. We examine how IoT-enabled sensor networks generate continuous, high-resolution data streams and how ML models transform these inputs into forecasts of pollution events, harmful algal blooms, microbial risks, and treatment performance. The review highlights major methodological trends, including the use of ensemble methods for classification, deep learning architectures such as LSTMs and CNNs for time series and image-based prediction, and emerging physics-informed and hybrid models that couple mechanistic insights with data-driven learning. At the architectural level, edge–fog–cloud frameworks dominate deployments, while communication protocols such as LoRaWAN, NB-IoT, and mesh networks are increasingly adapted to environmental monitoring. Despite promising advances, adoption remains limited by sensor reliability issues, data scarcity and imbalance, poor model generalization, lack of uncertainty quantification, cybersecurity vulnerabilities, and incomplete regulatory integration. We identify opportunities for progress through resilient sensor design, standardized open datasets, transfer learning, explainable AI, blockchain-enabled governance, and pathways to regulatory acceptance. By consolidating methods, applications, and future directions, this review positions IoT–ML systems as critical enablers of proactive, predictive water quality management. Addressing the technical and institutional gaps identified here will be essential for scaling these tools from pilots to operational frameworks capable of supporting sustainable water governance.