A review of machine learning applications in the prediction of selected groundwater quality parameters: Key lessons, knowledge gaps, and future directions

Rigorous prediction and assessment of groundwater quality is a critical component of sustainable water resource management. In recent years, machine learning (ML) has emerged as a powerful tool to model groundwater quality, offering data-driven solutions that can enhance the prediction reliability and broaden the scope of groundwater quality applications. This study presents a systematic and structured review of more than 100 peer-reviewed articles published over the last decade, identified through a comprehensive literature search and screening process using the Scopus database. The review protocol was designed to ensure coverage of high-quality, reproducible, and methodologically consistent studies, while enabling a transparent assessment of trends, gaps, and methodological practices in ML-based groundwater quality research. This review aims to evaluate the state-of-the-art application of ML in groundwater quality research through two complementary lenses: (i) the groundwater quality perspective, which focuses on groundwater parameters, watershed contexts, and scope of previous studies and (ii) the ML perspective, which considers types of ML models, ML features and target variables, dataset characteristics, ML model validation approaches, and evaluation metrics. The analysis highlights key outcomes and lessons, including the most used ML models, input and output variables, and evaluation metrics as well as the advantages of hybrid modeling approaches, the need for multi-parameter prediction, and the role of explainable AI in enhancing model generalizability and trust. Emerging opportunities, such as application of deep learning models and ML model generalization, were also identified as promising directions, while gaps remain in addressing under-studied regions, scaling across watershed sizes, and translating academic advances into practice. Overall, this review provides a structured roadmap to guide researchers and practitioners in applying ML for groundwater quality prediction, emphasizing both methodological rigor and practical impact. • A targeted review was done (>100 articles) on applying ML models in groundwater quality. • Random forest and neural network models showed superior predictive accuracy. • Nitrate was the most predicted groundwater quality parameter using ML models. • Small datasets (<250 points) were commonly used in ML modeling of groundwater quality. • Knowledge gaps include lack of interpretable ML models, deep learning, and hybrid modeling.