Towards intelligent air quality forecasting using integrated machine learning framework with variational mode...

Area of Science:

• Environmental Science
• Data Science
• Atmospheric Chemistry

Background:

• Accurate air quality (AQ) prediction is vital for public health and ecosystem well-being.
• Existing Machine Learning (ML) methods for AQ forecasting have limitations.
• Sulfur dioxide (SO₂) and nitrogen dioxide (NO₂) are key air pollutants requiring precise forecasting.

Purpose of the Study:

• To develop an efficient ML framework for daily SO₂ and NO₂ forecasting in Changping, China.
• To enhance prediction accuracy using integrated data decomposition and feature selection techniques.
• To evaluate the proposed model's performance against established ML approaches.

Main Methods:

• Utilized a Local Weights and Kernel Extreme Learning Machine (LWKELM) model.
• Integrated Catboost for efficient feature selection to identify influential input variables.
• Employed Multivariate Variational Mode Decomposition (MVMD) for input variable decomposition.
• Optimized model hyperparameters using the Interior Search Algorithm (ISA).

Main Results:

• The proposed MVMD-LWKELM-ISA model demonstrated superior performance.
• Achieved high accuracy for NO₂ (R=0.978, RMSE=0.537) and SO₂ (R=0.974, RMSE=1.965) forecasts.
• Outperformed Locally Weighted Linear Regression (LWLR), Gaussian Process Regression (GPR), KELM, and Multivariate Adaptive Regression Spline (MARS) models.

Conclusions:

• The developed framework provides highly accurate one-day-ahead forecasts for SO₂ and NO₂.
• The MVMD-LWKELM-ISA model is an effective and intelligent approach for daily air pollution forecasting.
• This research contributes to improved air quality management strategies through advanced ML.