Development and Evaluation of Statistical Models Based on Machine Learning Techniques for Estimating Particulate...

Area of Science:

• Environmental Science
• Data Science
• Atmospheric Chemistry

Background:

• Limited inter-country air quality models exist for Southeast Asia due to data scarcity.
• Accurate estimation of particulate matter (PM2.5 and PM10) is crucial for regional environmental health.

Purpose of the Study:

• To develop and evaluate machine learning models for estimating daily PM2.5 and PM10 concentrations in Brunei Darussalam.
• To assess the inter-country applicability of models trained in Singapore.

Main Methods:

• Utilized analysis of covariance (ANCOVA) and random forest regression (RFR) machine learning models.
• Trained models using air quality and meteorological data from Singapore.
• Validated models with data from Brunei Darussalam.

Main Results:

• The ANCOVA model demonstrated superior performance for PM2.5 (R2=0.94, RMSE=0.05 µg/m³)
• ANCOVA also showed better results for PM10 (R2=0.72, RMSE=0.09 µg/m³)
• Both models provided satisfactory PM concentration estimations, with ANCOVA excelling for concentrations over 18 µg/m³.

Conclusions:

• Machine learning models can effectively estimate PM concentrations across countries in Southeast Asia.
• The ANCOVA model shows significant potential for inter-country PM estimations.
• Further validation and data sharing can enhance regional air quality modeling.