Machine learning for spatiotemporal big data in air pollution

Abstract An accurate understanding of air pollutants in a continuous space-time domain is critical for meaningful assessment of the quantitative relationship between the adverse health effects and the concentrations of air pollutants. Traditional interpolation methods, including various statistic and nonstatistic regression models, typically involve restrictive assumptions regarding independence of observations and distributions of outcomes. Moreover, a set of relationships among variables need to be defined strictly in advance. Machine learning opens a new door to understand the air pollution data based on the exposing data-driven relationships and predicting outcomes without empirical models. In this chapter, the state-of-the-art machine learning methods will be introduced to unlock the full potential of the air pollutant data, that is, to estimate the PM2.5 concentration more accurately in the spatiotemporal domain. The methods can be extended to the other air pollutants.