Predicting Hourly Street-Scale NO2 and PM2.5 Concentrations Using Machine Learning at One of the Danish Traffic Hotspots

Road transport is one of the significant sources of ambient air pollution. Machine learning approaches are becoming increasingly popular to predict and forecast air quality. In conjunction, this study aims to explore the potential of commonly used machine learning algorithms, Artificial Neural Networks (ANN), Random Forests (RF), and Support Vector Machines (SVM) to accurately predict hourly, street-scale NO2 (ppb) and particulate matter (PM2.5) (µg/m3) at one of the Danish traffic hotspots, H. C. Andersens Boulevard (HCAB), in Copenhagen (2013–2015). Results suggest that the RF, overall, outperforms others [RF: RMSE = 11.2–13.5, R2 = 0.73–0.81; ANN: RMSE = 13.5–14.1, R2 = 0.69–0.75; SVM: RMSE = 14.7–15.1, R2 = 0.64–0.65]. One strength of this study is that it addresses the research gaps concerning the unexplored potential of machine learning techniques for air quality prediction in Denmark. Future work will include more predictor variables (e.g. traffic attributes) and testing of other techniques (e.g. Extreme Gradient Boosting) (XGBoost).