Machine learning-based evolution of water quality prediction model: An integrated robust framework for comparative...

Area of Science:

• Environmental Science
• Data Science
• Water Resource Management

Background:

• Accurate surface water quality prediction is crucial for sustainable resource management.
• Current deep learning models struggle with non-stationary environmental data and complex factor interactions.

Purpose of the Study:

• To introduce a novel, multi-level coupled machine learning framework for enhanced water quality prediction.
• To improve the accuracy and reliability of forecasting dynamic water quality indices.

Main Methods:

• Integration of data denoising, feature selection, and Long Short-Term Memory (LSTM) networks.
• Application of wavelet transform, moving average, and empirical mode decomposition techniques.
• Multi-step ahead predictions (t+1, t+3 days) with varying training data splits (80-20%, 70-30%).

Main Results:

• The LSTM model with data denoising improved prediction performance (R² increased by 1.01%).
• Wavelet transform integration showed superior adaptability, increasing R² by 0.81% and 0.51% over other methods.
• Model suitability varied based on time series variability patterns.

Conclusions:

• The integrated framework significantly enhances the prediction of dynamic water quality indices in complex settings.
• The proposed models demonstrate reliability and robustness under varying conditions.
• Further research is needed to validate the framework's scalability across diverse geographical and climatic conditions.