Forecasting ambient PM_2.5 and PM₁₀ concentrations in Hisar City through machine learning

Area of Science:

• Environmental Science
• Data Science
• Atmospheric Chemistry

Background:

• Hisar City faces significant PM2.5 and PM10 pollution from industrial, vehicular, and agricultural sources.
• Accurate particulate matter forecasting is crucial for effective air quality management strategies.

Purpose of the Study:

• To evaluate machine learning models for predicting PM2.5 and PM10 concentrations in Hisar City.
• To identify the most effective models for reliable air quality forecasting.

Main Methods:

• Collected historical air quality and meteorological data (2020-2024).
• Trained and tested models including Support Vector Regression, Random Forest, Gradient Boosting, AdaBoost, and M5.
• Utilized SHAP analysis for model explainability and sensitivity analysis of weather variables.
• Assessed performance using Correlation Coefficient (CC), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE).

Main Results:

• Ensemble models (Gradient Boosting and Random Forest) outperformed other regression models.
• Gradient Boosting achieved CC of 0.73 for PM2.5 and 0.668 for PM10, with minimal RMSE.
• Random Forest achieved CC of 0.724 for PM2.5 and 0.674 for PM10 during testing.

Conclusions:

• Machine learning techniques provide a reliable method for predicting air quality.
• Accurate PM forecasting supports policymakers in developing timely emergency response plans.