Predictive machine learning modeling of urban traffic air pollution

Abstract

Air pollution in Africa is a growing yet often overlooked threat, worsened by rapid industrialization, urban expansion, and road traffic. This study examines the impact of traffic flow and meteorological conditions on urban air quality along the Entebbe–Kampala corridor in Uganda. Traffic monitoring and air quality measurements were conducted using AirQo mobile sensors. Motorcycles, saloon cars, and light goods vehicles dominated traffic, while heavy trucks contributed minimally. Pearson correlation, regression, and XGBoost models analyzed pollutant relationships, with SHAP explaining variable contributions. Results showed PM₂.₅ peaks at night due to stable atmospheric conditions limiting dispersion. Regression analysis revealed strong positive relationships between Average Daily Traffic and PM₂.₅, CO₂, and TVOCs (R² = 0.89–0.99). Wind speed explained 45–71% of pollutant variance. The study underscores the need for integrated urban air quality policies focusing on traffic control, speed regulation, and fuel efficiency to mitigate emissions in rapidly urbanizing African cities.