par Mohamed Belalia (UQAM)
Résumé : In this talk a new methodology of estimating the predictive probability in a logistic regression setting is explored. The idea consists in writing this latter in terms of conditional copula and marginal distributions. The approach developing here constitute, first, selecting a parametric family of copula that describes better the data at hand, and estimating non parametrically the marginal distributions, sub- sequently, we use the plug-in method to build an estimate of the desired probability either in a binary case or multinomial one. The availability of a rich family of copula, a various goodness-of-fit test and a nonparametric estimation of the marginal makes the approach more flexible. The asymptotic properties related to these estimators are provided. Finally, a simulated study is carried out to evaluate the performance of the newly proposed procedure, besides the burn injury data are analyzed to show the flexibility of our method.
Joint work with Professors, Mhamed Mesfioui from UQTR and Taoufik Bouezmarni from Universit ́e de Sherbrooke.