Potential Assessment of Neural Network and Decision Tree Algorithms for Forecasting Ambient PM2.5 and CO Concentrations: Case Study

Abstract

Air pollution in megacities have caught attention of both researchers and policymakers because of increasing emissions, poor air quality, and potential adverse health impacts on densely inhabited populations. Oxides of nitrogen, particulate matter, carbon monoxide, and hydrocarbons are the major air pollutants of vehicular emissions near major intersections and arterials in megacities. The present study is mainly aimed at predicting PM2.5 and CO concentrations at an income tax office (ITO) intersection in the megacity of Delhi. Artificial neural networks (ANNs) and decision tree algorithms (e.g., REPTree and M5P algorithm techniques) are used to predict hourly fine particulate matter (PM2.5) and carbon monoxide (CO) pollutant concentrations at the ITO intersection. Factors and parameters, such as meteorological conditions, traffic, and vehicular emissions, that affect pollutant concentrations are used in different combinations for the model development. Performance evaluation of ANN, REPTree, and M5P algorithms for hourly PM2.5 and CO concentration prediction is carried out, and the effects of the aforementioned factors are discussed. The M5P algorithm performs better than ANN and REPTree algorithms in that it precisely captures the relationships among the predictor variables and pollutant concentrations.