Air Quality Prediction Using Edge-Gathered Traffic Data

Air pollution is a global health challenge and one of the leading causes of mortality with people from low- and middle-income countries being disproportionately affected. Traffic is one of the key contributors to urban air pollution and the ability to accurately model it is of primary importance to understanding its health implications. This paper presents the comparative studies of three AI-based predictive models such as Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM) network and a Gated Recurrent Unit (GRU) network to correctly identify the levels of air pollutants using traffic and environmental data. Eighteen months of weather, air quality and traffic data were used and the models were then tested on data without access to previous particulate matter (PM) 2.5 μm levels. The GRU provided the best balance of low training overheads with good performance. An autocorrelation analysis revealed that PM2.5 levels are highly autocorrelated, hence access to previous levels of PM2.5 is needed. The models created demonstrated that there is indeed a relationship between the variables.