Tree-Based Machine Learning Models for Classifying Safe and Unsafe Heavy Metal Levels in Groundwater: A Case Study from Jamshedpur Township, India

Tree-based machine learning (ML) models offer a powerful classification framework for rapidly screening groundwater for heavy metal contamination and associated health risks. This study applies several tree-based algorithms to classify groundwater samples from the Jamshedpur Township area, Jharkhand, India, as safe or unsafe with respect to selected heavy metals, using physicochemical parameters as predictors and WHO/BIS limits as class thresholds. Groundwater samples collected from shallow and deeper wells were analyzed for pH, EC, TDS, and heavy metals such as As, Pb, Cd, Cr, Ni, Cu, Zn, Fe and Mn, and compared with drinking water standards to define binary class labels. Groundwater samples were classified into safe and unsafe categories based on WHO/BIS standards and health risk thresholds (HI > 1, CR >10−4). Health risk assessment indicated significant non-carcinogenic and carcinogenic risks, particularly among children. Decision Tree, Random Forest, Gradient Boosting, and an Optimized Forest-type ensemble were trained and evaluated using accuracy, precision, recall, F1-score, and ROC–AUC, supported by confusion matrices. The Optimized Forest and Random Forest models yielded the highest classification performance, achieving high recall for unsafe samples, which is critical for public health screening, while feature importance analysis highlighted EC, TDS, pH, and specific ions as key predictors. The results indicate that tree-based ML models using routinely measured water quality parameters can serve as efficient decision-support tools for rapid identification of heavy metal risk zones in Jamshedpur Township and similar industrial urban environments.