Breathing Uneven Air: A Machine Learning Approach to Air Pollution in Chicago

Keywords: Air Pollution, Geospatial Analysis, PM2.5, r-INLA, Cook County
Abstract Type: Poster Abstract

Abstract

Air pollution is a major issue in urban areas, especially densely populated regions like Cook County--Chicago. This research focuses on predicting PM2.5 (Mean_PM2.5) levels by integrating geospatial data (r-INLA) and statistical modeling. We use ground-based air quality monitors and local sensors like PurpleAir to enhance prediction accuracy. Utilizing machine learning, specifically random forest models, we create detailed spatial maps of pollution patterns.
Incorporating factors like green spaces, weather conditions, and urban infrastructure helps us understand how pollution varies across neighborhoods. Inspired by Moraga’s geostatistical frameworks, we apply spatial prediction methods to account for the spatial correlation of PM2.5 across regions, allowing for more precise estimates in areas lacking direct measurements.
Our findings highlight disparities in pollution exposure, showing that socioeconomically disadvantaged communities face higher risks due to factors like traffic influence and limited green spaces. The resulting maps provide actionable insights for urban planners, public health officials, and policymakers to address pollution risks.
By combining random forest with geospatial analysis(r-INLA), this study advances real-time air quality forecasting and long-term environmental monitoring. It contributes to ongoing discussions on how geospatial data can address public health challenges, offering tools for cities to manage air pollution more effectively.