Application of machine learning algorithms to PM2.5 concentration analysis in the state of São Paulo, Brazil

Main Article Content

Angela Rosa Locateli Godoy
Ana Estela Antunes da Silva
Mirelle Candida Bueno
Simone Andréa Pozza
Guilherme Palermo Coelho


Air quality monitoring data are useful in different areas of research and have varied applications, especially with a focus on the relationship between air pollution, respiratory problems, and other health hazards. The main atmospheric pollutants are: ozone (O3), sulfur dioxide (SO2), carbon monoxide (CO), nitrogen dioxide (NO2), and particulate matter (PM). PM is one of the main objects of study when one intends to protect people from exposure to pollutants. This study contributes to the analysis of PM2.5 in 21 stations in the state of São Paulo monitored by the Environmental Company of São Paulo State (CETESB). It employs cluster analysis, a prominent data mining method for detecting patterns and discovering similarities which is important for assessing air pollution, especially in a geographically vast area such as that of the state of São Paulo, which does not follow a single pattern. Another data mining technique (association rules) supports the analysis of the relationship between pollutants and meteorological variables, as it allows identifying changes between elements that occur together, in a wide variety of data. Our objectives include determining stations with similar behaviors and exploring the temporal variety of the pollutant as it relates to the dominant meteorological factors in the periods of high concentration. The clustering algorithm automatically separates stations according to their monthly averages of PM2.5 concentration between 2017 and 2019. The clusters of stations that showed the highest pollution rates essentially included urban centers with emissions by industries and vehicles, while those with the lowest rates were located further inland. A cyclical behavior in pollutant variation was also observed in the three years under study and for both clusters. For the months with the highest concentration of PM2.5, association rule learning was applied to connect air temperature, relative humidity, and wind speed with PM2.5 and carbon monoxide (CO) concentrations. The obtained results are useful to analyze the temporal and geolocation profiles of pollution by particulate matter, since they identify the behavior of the meteorological factors that predominate in periods of greater concentration.

Article Details

How to Cite
Godoy, A., Silva, A., Bueno, M., Pozza, S., & Coelho, G. (2021). Application of machine learning algorithms to PM2.5 concentration analysis in the state of São Paulo, Brazil. Brazilian Journal of Environmental Sciences (Online), 56(1), 152-165. Retrieved from