Time series-based PM_2.5 concentration prediction in Jing-Jin-Ji area using machine learning algorithm models.

Area of Science:

• Environmental Science
• Data Science
• Public Health

Background:

• Air pollution, particularly PM2.5 concentration, is a global health concern affecting mortality rates.
• Machine learning (ML) models are increasingly used for air pollution prediction, outperforming traditional methods.
• Limited research exists on ML model selection and interpretation for environmental policy.

Purpose of the Study:

• To compare the performance of four ML algorithms (LinearSVR, K-Nearest Neighbor, Lasso, Gradient Boosting) for PM2.5 prediction.
• To evaluate model accuracy across different cities and seasons.
• To provide insights for environmental policy making regarding air pollution control.

Main Methods:

• Utilized historical five-day data to forecast the next day's PM2.5 concentration.
• Compared LinearSVR, K-Nearest Neighbor, Lasso regression, and Gradient Boosting algorithms.
• Analyzed model performance based on Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) at city and seasonal levels.

Main Results:

• ML models demonstrated accuracy in predicting next-day PM2.5 concentrations.
• Gradient Boosting showed superior performance at the city level, with MAE of 9 ug/m³ and RMSE of 10.25-16.76 ug/m³.
• All models performed best in winter and worst in summer, highlighting seasonal variations.

Conclusions:

• ML models are effective for short-term PM2.5 forecasting.
• Gradient Boosting is a promising model for city-specific air quality prediction.
• Understanding seasonal and city-specific model performance is vital for targeted environmental policy interventions.