A Review on Predictive AI Models for Water Quality in Chilika River, Odisha: High-Precision Parameter Forecasting, Contaminant Detection, and Data-Driven Sustainability Strategies

AbstractPrecise evaluation and prediction of water quality are very important for sustainable watersupply control and public health protection, particularly in the context of rapid urbanization,climate variability, and increasing anthropogenic pressures on freshwater resources. MachineLearning (ML) and Deep Learning (DL) methods have proven to be influential tools formodelling the spatiotemporal, multifaceted, nonlinear, and performance aspects of watersystems that are often difficult to capture using traditional statistical or process-based models.This review systematically studies the state of science of ML and DL methods, with SupportVector Machines, Random Forests, Gradient Boosting, Convolutional Neural Networks(CNNs), Long Short-Term Memory (LSTM) networks, and their hybrid combinations, such asCNN–LSTM architectures which effectively integrate spatial feature extraction with temporalsequence learning. The study highlights their performance in predicting key water qualityparameters, such as pH, dissolved oxygen, and ammonium Nitrate, which are widelyrecognized as critical indicators of aquatic ecosystem health, pollution levels, and regulatorycompliance, across various water sources, including rivers and lakes. Comparative analysesreported in the literature indicate that ensemble-based ML models often provide robust baselineperformance, while DL models excel in complex, high-dimensional scenarios with sufficientdata availability. Despite notable advancements in predictive accuracy, existing modelscontinue to face persistent challenges, including limited generalization across diverse datasets,data sparsity, sensor fouling, and a lack of model interpretability, which collectively hindertheir scalability, reliability, and acceptance by domain experts and decision-makers. Thisreview identifies critical research gaps and outlines future directions, emphasizing theintegration of Explainable AI wavelet-based denoising and IoT-enabled monitoringframeworks to improve noise resilience, enhance trust, and support continuous real-time dataacquisition for real-time, adaptive, and transparent water quality management. Theincorporation of explainability mechanisms is particularly important for understanding modelbehaviour, supporting regulatory decisions, and enabling stakeholder confidence. The resultshighlight the possibility of hybrid ML/DL frameworks to transform conventional water qualityanalysis, including AI-based analysis, into intelligent, data-driven decision support toolscapable of early warning, proactive intervention, and long-term sustainability planning forwater quality management.