Comparative Analysis of Machine Learning Models in Groundwater Quality Assessment: A Systematic Review

Groundwater Quality (GWQ) presents a significant global concern owing to extensive agricultural and industrial activities, necessitating proficient management strategies, as it constitutes approximately half of the world's potable water supply. Traditional methods of monitoring water quality are valuable for identifying pollution sources, but they do not adequately provide an overall view of water quality trends. Machine learning (ML) technologies provide reliable predictions of water quality by learning complicated patterns from data without preconceived equations. This paper provides a meta-analysis and bibliographic review of ML applications in GWQ assessment, evaluating the accuracy, applicability, and usability of various models including artificial neural networks (ANN), support vector machines (SVM), k-nearest neighbors (KNN), decision tree (DT), random forest (RF), and deep learning (DL) (including deep neural networks [DNN] architectures) approaches. It offers a thorough synopsis and conclusion that neural networks have traditionally been the most widely used ML model in GWQ modelling. India, China, and the United States are global leaders in groundwater modelling, benefiting from extensive historical data. The most widely modelled elements are nitrate and heavy metal pollution, which are present in about half of the studies. Despite significant progress, several research gaps remain, particularly in the modelling of lesser-explored water quality parameters and in the integration of advanced ML techniques. Emerging approaches such as physics-informed machine learning (PIML), graph neural networks (GNNs), transformer-based architectures, and large language models (LLMs) show considerable potential for improving prediction accuracy and handling complex environmental datasets. These advancements may enable more robust and integrated GWQ management frameworks in future studies.