Applying machine learning methods in managing urban concentrations of traffic-related particulate matter
This study presents a new method for evaluating the effectiveness of roadside PM10 and PM2.5 reduction scenarios using Machine Learning (ML) based models. The ML methods include Artificial Neural Networks (ANN), Boosted Regression Trees (BRT) and Support Vector Machines (SVM). Traffic, meteorological and pollutant data collected at nineteen Air Quality Monitoring (AQM) sites in London for a period between 2007 and 2012 was used. The ML models performed very well in predicting the concentrations of PM10 and PM2.5 with around 95% of their predictions falling within the factor of two of the observed concentrations at the roadsides. The prediction errors observed were very small as indicated by the average normalised mean gross errors of 0.2. Also, the predictions of the models correlated well with the observed concentrations as shown by the average values of R (0.8) and index of agreement (0.74). Additionally, when some PM10 and PM2.5 reduction scenarios were modelled, the ML models predicted various degree of reductions in the roadside concentrations. In conclusion, well trained ANN and BRT models can be successfully applied in predictions of roadside PM10 and PM2.5 concentrations. Moreover, they can be applied in measuring the effectiveness of roadside particle reduction scenarios.
Machine learning, Air quality, Particulate matter, Traffic emissions, Neural networks, Boosted regression trees