Accurate and efficient prediction of atmospheric PM₁, PM_2.5, PM₁₀, and O₃ concentrations using a customized software...

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• Particulate matter (PM) and ozone (O3) are significant air pollutants with adverse health effects.
• Accurate and efficient prediction of PM and O3 concentrations remains a challenge.
• These pollutants are secondary in nature, complicating forecasting.

Purpose of the Study:

• To develop and evaluate an eXtreme Gradient Boosting (XGBoost) model for predicting PM (PM1, PM2.5, PM10) and O3 concentrations.
• To investigate the impact of various datasets, features, and model parameters on prediction accuracy.
• To create a user-friendly software application for air pollutant concentration prediction.

Main Methods:

• Utilized one-year of monitoring data including PM (PM1, PM2.5, PM10), O3, meteorological parameters, and precursor concentrations (NOx, SO2, CO, alkanes, aldehydes, ketones).
• Employed the eXtreme Gradient Boosting (XGBoost) machine learning model for prediction.
• Applied Principal Component Analysis (PCA) for feature dimensionality reduction.
• Developed a graphical user interface (GUI) using PyQt5 for the prediction application.

Main Results:

• The XGBoost model demonstrated efficient and accurate predictions for PM1, PM2.5, PM10, and O3 concentrations.
• Feature selection and dimensionality reduction using PCA improved prediction efficiency.
• The developed software enables simultaneous prediction of all four pollutants.

Conclusions:

• The XGBoost model provides a robust and efficient approach for predicting key air pollutant concentrations.
• The integrated software application enhances the convenience and accessibility of air quality forecasting.
• This work contributes to improved air quality monitoring and management strategies.