Surface water quality index forecasting using multivariate complementing approach reinforced with locally weighted...

Area of Science:

• Environmental Science
• Water Quality Management
• Data Science & Machine Learning

Background:

• River water quality monitoring is crucial for community health and environmental protection.
• Accurate forecasting of biochemical oxygen demand (BOD) is vital but challenging with traditional methods.
• Existing methods lack the accuracy and long-term reliability needed for effective BOD detection.

Purpose of the Study:

• To introduce and evaluate an innovative hybrid model (MVMD-LWLR) for predicting 5-day BOD levels.
• To enhance BOD forecasting accuracy in the Klang River, Malaysia, using advanced computational techniques.
• To compare the proposed model's performance against established regression and machine learning algorithms.

Main Methods:

• Developed a hybrid model combining Multivariate Variational Mode Decomposition (MVMD) and Locally Weighted Linear Regression (LWLR).
• Employed Categorical Boosting (Catboost) for feature selection to identify significant input variables.
• Utilized Gradient-Based Optimization (GBO) for fine-tuning model parameters and enhancing predictive accuracy.

Main Results:

• The MVMD-LWLR model demonstrated superior performance in forecasting BOD compared to Kernel Ridge, LASSO, Elastic Net, and Gaussian Process Regression.
• Optimized MVMD-LWLR with GBO achieved higher accuracy and minimal error in BOD predictions.
• Evaluation metrics including RMSE, R, U_95%, and NSE confirmed the model's robustness and reliability.

Conclusions:

• The proposed MVMD-LWLR hybrid model offers a significant advancement in river water BOD forecasting.
• This optimized approach provides reliable and accurate predictions essential for effective water quality management.
• The study highlights the potential of integrating advanced decomposition, regression, and optimization techniques for environmental monitoring.