Comparative Evaluation of Machine Learning Models for Spatio‐Temporal Groundwater Quality Prediction and Classification in the Muvattupuzha River Basin, India

Groundwater contamination has increased significantly due to population growth, land-use change, and unsustainable resource exploitation, necessitating advanced predictive tools for effective water governance. This study presents a multi-temporal, comparative machine learning (ML) to evaluate groundwater quality in the Muvattupuzha River Basin, Kerala, India, using datasets from 2003, 2013, and 2023. A total of five supervised ML models (i.e., decision tree [DT], logistic regression, support vector machine, random forest, and k-nearest neighbor) are systematically assessed to distinguish groundwater as safe or unsafe. Model performance is evaluated using accuracy, recall, F1-score, coefficient of determination, and root-mean-square error. Among the five models, the DT consistently outperforms others, achieving a maximum classification accuracy (96%). It also demonstrates strong interpretability under data-limited conditions. The novelty of the presented work lies in integrating model-specific feature importance with hydrochemical reasoning, revealing that salinity-related parameters serve as effective surrogate indicators for large-scale groundwater quality screening, while nutrients and hardness reflect localized anthropogenic and geogenic controls. The temporal analysis captures the evolving dynamics of groundwater quality over two decades. It is highlighting emerging risks despite partial improvements. Overall, the proposed model advances interpretable, data-driven groundwater assessment and provides actionable insights for early warning, sustainable monitoring, and policy-oriented water resource management in rapidly transforming river basins.