Traffic NOx Pollution Prediction and Health Cost Estimation Using Machine Learning: A Case Study of Toronto, Canada

Road traffic is a significant source of air pollution that has a harmful impact on human health. To reduce the health and environmental impacts of fossil fuel consumption in the transportation sector, many countries have implemented policies to promote the deployment of electric vehicles (EVs). A vital factor to consider when designing policies to support EV use is the monetized health impacts of fossil fuel consumption. This research aims to investigate the health benefit of replacing internal combustion engine vehicles (ICEVs) with zero-emission vehicles in the city of Toronto, Canada. A long short-term memory (LSTM) model is developed in this work to predict future NOx concentrations considering the effect of the traffic volume, weather, time of day, and historical NOx concentrations. The developed model is then used to predict long-term NOx concentrations and annual average NOx reduction from zero-emission vehicle deployment in four different scenarios in Toronto. Additionally, interpolation methods are used to predict the pollution reduction in all Dissemination Areas (DA) of Toronto, and a health cost assessment is conducted to estimate the health benefit in all the scenarios. The results of the modeling in this work show that the western areas of Toronto experience higher NOx concentration reduction in all scenarios. These reductions are the result of the higher correlation between traffic volume and pollution in those areas. The results also show that with a 10% reduction in ICEV traffic volume, 70 premature deaths can be prevented annually, equivalent to 560 million CAD in health benefits per year.