Predicting water quality index using stacked ensemble regression and SHAP based explainable artificial intelligence.

Area of Science:

• Environmental Science
• Data Science
• Water Resource Management

Background:

• Accurate Water Quality Index (WQI) forecasting is crucial for effective water resource management and public health protection.
• Existing forecasting methods often lack interpretability and high predictive accuracy.
• Physicochemical parameters are key indicators of river water quality.

Purpose of the Study:

• To develop and validate a novel stacked regression ensemble model for WQI forecasting.
• To integrate Explainable Artificial Intelligence (XAI) for model interpretability.
• To identify the most influential physicochemical parameters for WQI prediction.

Main Methods:

• A stacked ensemble model was developed using six machine learning algorithms (XGBoost, CatBoost, Random Forest, Gradient Boosting, Extra Trees, AdaBoost) with Linear Regression as the meta-learner.
• The model was trained on 1,987 Indian river water quality samples (2005-2014) using seven normalized physicochemical parameters.
• SHAP (Shapley Additive explanations) was employed for model interpretability and feature importance analysis.

Main Results:

• The stacked ensemble model achieved superior performance with R² of 0.9952, Adjusted R² of 0.9947, MAE of 0.7637, and RMSE of 1.0704.
• Individual models CatBoost and Gradient Boosting showed strong standalone performance.
• SHAP analysis identified Dissolved Oxygen (DO), Biochemical Oxygen Demand (BOD), conductivity, and pH as the most influential parameters for WQI prediction.

Conclusions:

• The proposed stacked ensemble model with XAI offers high predictive accuracy and interpretability for WQI forecasting.
• This approach enhances real-time environmental monitoring and supports automated policy frameworks.
• The findings build stakeholder confidence in water resource sustainability through improved forecasting and transparency.