Laboratoire de Mathématiques de Besançon - UMR 6623 CNRS

Accueil > Agenda scientifique

17 février 2020: 1 événement

  • Planning des séminaires 2019-2020

    Lundi 17 février 11:00-12:00 - Jasper Velthoen - EWI - TU-Delft, Pays-Bas

    Séminaire PS : Using interpretable random forests to forecast maximum temperature

    Résumé : In the past years random forest has been picked up by the meteorological community. There it has been used in the context of post-processing, i.e. estimating the statistical relation between observations and forecasts in order to correct systematic biases in the forecasts. The main argument for random forests has been its ability to deal with a high dimensional covariate space. On the other hand, a disadvantage of such a model is that it becomes hard to interpret. This becomes especially a problem in the presence of high correlations between the covariates, which are observed in weather forecasting. Interpretation is essential as the forecasts need to be used and interpreted by practitioners.
    We introduce a forward variable selection method that will allow us to select a small set of predictive variables whilst keeping the same predictive power. By reducing the number of covariates, we show that it is much easier to interpret the random forest and analyse the effect of the covariates on the predictive distribution.

    Lieu : Salle 316B - 16 route de gray
    25030 Besançon cedex

    En savoir plus : Planning des séminaires 2019-2020