Jeudi 21 Novembre


Retour à la vue des calendrier
Jeudi 21 Novembre
Heure: 12:15 - 13:30
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Classification with reject option
Description: Blaise Hanczar Given a classification task, an approach to improve accuracy relies on theuse of classifiers with reject option. These classifiers are trained toreject observations for which predicted values are not reliable enough.During the classifier construction, a rejection area is defined around thedecision boundary in the feature space. There are two approaches fordefining the reject option: the plug-in rules where two decision thresholdsare computed on the classifier output and the embedded rejection rules wherethe rejection area is computed during the classifier fitting. We propose newmethods of plug-in reject for small-sample data and new ways of research forthe embedded reject.