Automated machine learning achieves accurate water quality prediction with reduced parameter requirements.

Area of Science:

• Environmental Science
• Data Science
• Water Resource Management

Background:

• Accurate water quality assessment is vital for environmental monitoring and public health.
• Traditional Water Quality Index (WQI) methods are often complex and time-consuming.
• Machine Learning (ML) offers potential but requires significant expertise and resources.

Purpose of the Study:

• To evaluate AutoML's predictive performance for WQI using a reduced parameter set.
• To assess model interpretability through feature importance analysis.
• To propose an automated framework for efficient water quality monitoring.

Main Methods:

• Utilized the AutoGluon platform for Automated Machine Learning (AutoML).
• Analyzed a 36-year dataset from Taiwan's national river water quality monitoring network.
• Focused on four key parameters: electrical conductivity (EC), suspended solids (SS), water temperature (WT), and pH.
• Employed an automated pipeline for model selection, hyperparameter tuning, and ensemble construction.
• Evaluated ensemble-based tree models including CatBoost, Random Forest, and XGBoost.

Main Results:

• AutoML models achieved a mean R-squared of 0.76 with low predictive errors.
• Ensemble tree models demonstrated superior performance in predicting WQI.
• Electrical conductivity (EC) was identified as the most significant predictor.
• Reduced input parameters did not compromise prediction accuracy.

Conclusions:

• AutoML provides a feasible and efficient method for water quality assessment.
• Fewer input parameters can yield accurate WQI predictions, reducing costs and time.
• The developed automated framework supports rapid and reliable environmental monitoring.