Robust machine learning algorithms for predicting coastal water quality index.

Area of Science:

• Environmental Science
• Water Quality Management
• Machine Learning Applications

Background:

• Coastal water quality is crucial for ecosystem health.
• Traditional Water Quality Index (WQI) methods face reliability issues.
• Machine learning (ML) offers potential for more robust WQI prediction.

Purpose of the Study:

• To identify reliable ML algorithms for predicting coastal Water Quality Index (WQI).
• To reduce model uncertainty in WQI calculations for Cork Harbour.
• To compare the performance of eight common ML algorithms.

Main Methods:

• Applied eight ML algorithms: RF, DT, KNN, XGB, ExT, SVM, LR, GNB.
• Utilized a 70% training and 30% testing data split.
• Validated models using 10-fold cross-validation and evaluated with RMSE, MSE, MAE, R², and PREI metrics.

Main Results:

• Decision Tree (DT), Extra Tree (ExT), and Extreme Gradient Boosting (XGB) demonstrated superior performance with near-perfect scores (RMSE=0.0, R²=1.0).
• Random Forest (RF) also showed strong predictive capability (R²=0.98).
• DT, ExT, and XGB models significantly reduced uncertainty in WQI predictions.

Conclusions:

• DT, ExT, and XGB are effective and robust ML models for predicting coastal WQI.
• These algorithms offer a significant improvement over traditional WQI methods by reducing model uncertainty.
• Findings support optimizing WQI model architecture for enhanced water quality monitoring.