Comparative Analysis of Traditional and Ensemble Models for Water Quality Index Prediction with Explainable AI

Abstract

Accurate prediction of water quality is vital for effective environmental management. This study presents a comparative analysis of traditional and ensemble machine-learning models for predicting the Canadian Council of Ministers of the Environment Water Quality Index (CCME-WQI) using EPA Ireland coastal monitoring data. A standardized and leakage-proof pipeline was employed with robust scaling and multiple cross-validation across multiple random seeds to ensure stable and reproducible performance. Among all models, XGBoost achieved the best performance (R² = 0.991). Model interpretability was enabled by SHAP analysis supported by feature correlation that identified Dissolved Oxygen as the dominant factor of WQI. Overall, results illustrate the potential of ensemble learners combined with explainable AI in making accurate, interpretable, and generalizable water-quality predictions to enable data-driven environmental monitoring and decision-making.