Developing river water quality prediction model incorporating reliable indexing approach

Area of Science:

• Environmental Science
• Data Science
• Water Resource Management

Background:

• Hyperparameter optimization is crucial for machine learning (ML) model performance.
• Accurate water quality (WQ) prediction is essential for sustainable water resource management.
• Data-driven models offer a promising approach for WQ assessment.

Purpose of the Study:

• To evaluate five hyperparameter optimization techniques for ML-based WQ prediction models.
• To compare the performance of ten ML algorithms and fifty models for predicting WQ index (WQI) scores.
• To identify the most effective optimization technique and ML model for WQ prediction using the RMS-WQI approach.

Main Methods:

• Utilized ten ML algorithms and fifty models for WQ prediction.
• Employed five hyperparameter optimization techniques, including OPTUNA.
• Evaluated models using RMSE, MSE, MAE, PABE, NSE, and R² for sensitivity analysis.

Main Results:

• The Gradient Boosting Regression (GBR) model optimized with OPTUNA (GBR-OPT) demonstrated superior performance in both training and testing phases.
• GBR-OPT achieved the lowest RMSE (0.45) and highest efficiency (NSE = 0.71) for WQ prediction.
• The GBR-OPT model exhibited high sensitivity with R² values of 0.99 and 0.98 for 2021 and 2022, respectively.

Conclusions:

• Hyperparameter optimization significantly impacts the reliability of WQ prediction models.
• The GBR-OPT model is highly effective for accurate and efficient WQ index prediction.
• The findings support the development of advanced WQ prediction models for sustainable water management.