Combining machine learning models through multiple data division methods for PM_2.5 forecasting in Northern Xinjiang,...

Area of Science:

• Environmental Science
• Atmospheric Science
• Data Science

Background:

• Particulate Matter (PM2.5) poses significant air quality challenges.
• Accurate forecasting of PM2.5 is crucial for public health and environmental management.
• Previous forecasting models often struggle with the complex, nonlinear dynamics of air pollution.

Purpose of the Study:

• To develop and evaluate hybrid models for daily average PM2.5 forecasting in Northern Xinjiang, China.
• To assess the effectiveness of combining different modeling approaches for improved prediction accuracy.
• To investigate the influence of meteorological and air pollutant data on PM2.5 forecasting.

Main Methods:

• Developed hybrid models by combining Back Propagation Artificial Neural Network (BPANN) and Multiple Linear Regression (MLR).
• Utilized daily meteorological and air pollutant data from 2015-2019 as input variables.
• Evaluated model performance using Leave-One-Out Cross-Validation (LOOCV), fivefold cross-validation, and hold-out methods.

Main Results:

• Combined models generally outperformed individual member models (BPANN and MLR).
• The optimal combined model achieved high correlation coefficients (R ≈ 0.87 in January, R ≈ 0.946 in August).
• Forecasting performance was better in August compared to January for both member and combined models.

Conclusions:

• Combining linear (MLR) and nonlinear (BPANN) models is an effective strategy for PM2.5 forecasting.
• Multiple data division methods enhance model evaluation and robustness.
• The developed hybrid models offer a promising tool for air quality management in the region.