Air quality in Bogotá, Colombia, has become of increasing concern. Especially, the levels of PM <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">10</sub> are alarming, because of their relation to health risks. A forecast system for PM <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">10</sub> levels is beneficial for developing preventive policies of environmental authorities. This paper proposes different forecasting models of particulate matter obtained with three machine learning techniques. A dataset from 8 air quality monitoring stations including PM <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">10</sub> and environmental measurements was constructed. Three selection methods of relevant variables for prediction were assessed: selecting variables with the assistance of an expert group, and using two automatic selection methods. Having three sets of potential variables to use as an input, three different forecasting methods were implemented: logistic regression, classification trees and random forest. Finally, a validation and comparison of results are made, to conclude about the best forecast model to be implemented for the city.