Comparative Analysis of Random Forest and Decision Tree for Arsenic Prediction in Groundwater in Begusarai, Bihar

A comprehensive statistical analysis was conducted on groundwater samples from Matihani and Cheria Bariarpur blocks of Begusarai District, Bihar, to assess physicochemical properties and heavy metal concentrations. While pH levels in both blocks remained within acceptable limits (6.5–8.5), arsenic (As) concentrations significantly exceeded the permissible limit (0.01 µg/L), with mean values of 0.129 µg/L in Matihani and 0.1656 µg/L in Cheria Bariarpur. Machine learning models—Decision Tree Regressor (DTR) and Random Forest Regressor (RFR)—were applied to predict arsenic concentrations. RFR outperformed DTR, achieving a lower MSE (0.001867), RMSE (0.043208), and higher NSE and R² values (0.895), demonstrating superior predictive accuracy. The results indicate that arsenic contamination poses a serious public health risk, and Random Forest offers a robust tool for spatial prediction and water quality management. This study highlights the importance of integrating AI-driven modeling with field data to enhance groundwater risk assessment and support informed mitigation strategies.