Assessment of Heavy Metal Contamination in Groundwater Coupled with Integrated Machine Learning-Based Arsenic Prediction

Monitoring groundwater quality is critical for ensuring public health, environmental sustainability, and effective water resource management. Traditional monitoring methods, though reliable, are often time-consuming and lack predictive capabilities. This study applies machine learning (ML) models—Decision Tree Regressor (DTR) and Random Forest Regressor (RFR)—to predict arsenic concentration in groundwater samples collected from Bodh Gaya, Bihar, India, during the pre-monsoon season. Fifty samples were analyzed for pH and heavy metals, revealing frequent exceedance of permissible limits for Fe, Al, Mn, and As. Data preprocessing included normalization, scaling, and splitting into training and testing sets. Model performance was evaluated using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Coefficient of Determination (R²), and Nash–Sutcliffe Efficiency (NSE). Results indicate that DTR outperformed RFR with lower error values (MSE = 0.00210, RMSE = 0.04365) and higher accuracy (NSE = 0.891, R² = 0.891). Feature importance analysis identified chloride as the most influential predictor, with varying contributions from other metals. The findings demonstrate the potential of ML-based approaches for real-time water quality prediction, enabling proactive interventions for sustainable groundwater management.