Modeling Particulate Matter Concentrations: A Comparative Analysis of Statistical and Machine Learning Models

Abstract

This study compares Ordinary Least Squares (OLS) regression and Random Forest models to analyze particulate matter (PM1, PM2.5, and PM10) concentrations based on meteorological and traffic data collected on major arterials in Karachi, Pakistan. OLS regression highlights temperature and humidity as significant contributors to PM levels, while wind speed shows an inverse relationship, especially with PM1. Random Forest regression demonstrates superior performance with higher R2 values and a lower RMSE, effectively capturing complex, non-linear relationships among variables. Wind speed thresholds for PM dispersion are identified, providing critical benchmarks for air quality management. This comparative analysis underscores the effectiveness of machine learning methods for accurate environmental modeling.