Prediction of particulate matter concentration in urban environment using Random Forest

Particulate matter (PM) is a major air pollutant that can have adverse effects on human health, especially for vulnerable populations such as children, the elderly, and those with respiratory or cardiovascular conditions. This study presents a method for prediction of particulate matter concentration with aerodynamic diameter smaller then 10 μm (PM10) in an urban environment. Meteorological data and vehicle flow data from an urban road in Curitiba, Brazil, were used. The air quality was analyzed in two monitoring points located 1 km apart, the sampling points are named Politécnico and Perkons, where SDS011 optical sensors were installed. The prediction was based on the machine learning algorithm Random Forest (RF). The baseline concentration was a dataset from historical records of particulate matter measurements from monitoring stations in Curitiba. Several scenarios were tested and it was concluded that the daily time scale presents the best performance in PM10 prediction, with 80.42% accuracy, using the baseline and PM10 Perkons as descriptors. The most important meteorological variables for the prediction were: air temperature (°C), wind speed (m/s), and wind gust (m/s). Throughout the day there were two peaks with large amounts of pollutants in the air, near 8:00 am and 6:00 pm, times when there are the largest flows of vehicles circulating on the road. The Random Forest algorithm proved to be a good estimator of PM concentration, which is a proxy for air pollution.