Precise estimation of water quality is essential for managing water pollution effectively and improving water management practices. Although, numerous recent investigations have documented inconsistencies in conventional water quality indexing methods owing to inherent subjective nature of its computation. Therefore, this study introduced a novel approach that integrates ENTROPY and CRITIC methods and applied sensitivity analysis-driven machine learning (ML) algorithms for groundwater quality prediction. The aim was to enhance prediction accuracy and robustness while gaining valuable insights into hydrochemical processes influencing groundwater quality. Six ML models, i.e., Support Vector Machine (SVM), Artificial Neural Network (ANN), Linear Regression Model (LRM), Regression Tree (RT), Gaussian Process Regression (GPR), and Boosted Regression Tree (BRT) were applied to forecast water quality across Tamil Nadu, India. This study addresses a significant gap, as systematic ML framework has been lacked to evaluate and predict drinking water quality across the same state, covering 619 monitoring sites. Calibration of the proposed Integrated Water Quality Index (IWQI) models was conducted using 75 % of the dataset (465 training wells) and validated with 25 % (154 monitoring wells), considering 13 physicochemical parameters. Rigorous statistical evaluations employing metrics such as RMSE, MSE, and MAE were performed to identify most effective prediction model, accounting for both uncertainty and predictive performance. This study suggest that ANN model is the most suitable, exhibiting the lowest uncertainty with R2 = 1 and the best predictive performance statistics, including RMSE (training = 0.0264, testing = 0.5638), MSE (training = 0.0007, testing = 0.3179), and MAE (training = 0.0083, testing = 0.0729). Sensitivity analysis identified electrical conductivity (EC) as highest responsive input parameter, while fluoride (F−) showed the lowest sensitivity. The robustness of the ANN model was evident, maintaining accuracy even with perturbations up to 10 %, thereby enabling precise IWQI score forecasting across all monitoring stations.