Air Pollution Prediction using Machine Learning: A Review

Abstract

In the effort to achieve accurate air pollution predictions, researchers have contributedvarious methodologies with varying data and different approaches that can be judgedaccurate in their respective contexts. Diverse approaches have been used so far in theliterature to achieve optimal accuracy in the prediction of air pollution. Researchers havealso used different combinations of data such as Meteorological, Traffic and Air Qualitydata. Hence, creating a situation where there are open questions on which of the machinelearning (ML) algorithms or ensemble of algorithms is best suited for various combinationsof data and varying dependent and independent variables. While it is obvious that there isa need for a more optimally performing predictive model for air pollution prediction, it isdifficult to know what combination of algorithms and data is best suited for variousdependent variables. In this study, we reviewed 26 research articles reported recently in theliterature and the methods applied to different data to identify what combination of MLalgorithms and data works best for the prediction of various air pollutants. The studyrevealed that despite the availability of many datasets, researchers in this domain cannotavoid the use of Air Quality and Meteorological datasets. However, Random Forest appearsto perform well for various combinations of datasets.