In the scientific literature, numerous studies show that meteorological conditions have an impact on the emission, dispersion and suspension of pollens in the air. Several allergenic species permanently threaten the health of millions of people in France. Preventive information on the risk of pollen exposure would become a real asset for allergy sufferers. The main objective of this article is to study, thanks to statistical learning techniques using historical data and meteorological parameters of the day (D), the ability to predict 3 days (D+3) in advance the risk levels of pollen presence in the air on a given territory (in Metropolitan France). We were interested in the prediction -in 4 levels- of risk for 3 families of pollens which are among the most allergenic species (Ambrosia, Cupressaceae and grasses). The aggregation of binary logistic regression models for each level of risk by a random forest classifier allowed us to predict the level of pollen risk with performances in the range of 75% to 90% of AUC and 70% of precision and recall, confusions concerning mainly low and medium levels.