Deep Learning for Water Quality Index Prediction: A Focus on Integrating Biological Parameters-A Systematic Review.

This systematic review synthesizes deep learning applications for predicting the Water Quality Index (WQI), with particular emphasis on integrating biological indicators. Following PRISMA 2020 guidelines, a comprehensive search was conducted across Scopus, Web of Science, IEEE Xplore, ScienceDirect, and Google Scholar for studies published between 2015 and 2026. From 312 initial records, 50 studies met the inclusion criteria after duplicate removal and title, abstract, and full-text screening. The review reveals that advanced architectures such as CNNs, LSTMs, BiLSTMs, GRUs, Transformers, and hybrid CNN-LSTM models increasingly outperform traditional statistical and machine learning approaches by effectively capturing nonlinear and spatiotemporal water-quality dynamics. However, only 18% of the reviewed studies incorporated biological indicators such as biochemical oxygen demand (BOD), fecal coliforms, Escherichia coli, or chlorophyll-a, and fewer than 8% explicitly considered microbial contamination variables. Studies integrating biological parameters consistently reported lower prediction errors, with error reductions of 22 to 12% in hybrid models, and improved contamination-risk detection compared with physicochemical-only approaches. Key challenges include limited biological monitoring data, high laboratory costs, inconsistent sampling frequency, lack of standardized benchmark datasets, poor model interpretability, and data scarcity, particularly in developing regions such as Sub-Saharan Africa. The review also identifies significant geographical imbalances, with most studies originating from regions with advanced monitoring infrastructure. To address these gaps, the conceptual Bio-Enhanced Water Quality Index (BE-WQI) framework is proposed, integrating physicochemical and biological indicators within a hybrid CNN-LSTM architecture combined with explainable artificial intelligence (XAI) techniques such as SHAP to enhance predictive accuracy, transparency, and health-oriented decision-making. This framework offers a practical pathway to improve WQI forecasting, especially in data-scarce, resource-constrained environments. Future research should prioritize standardized open-access datasets, cost-effective biosensors, transfer learning, and biologically informed deep learning models to support sustainable water resource management and public health protection.