Méthodes d’apprentissage faiblement supervisé : arbres de décision et classification itérative

This paper addresses the probabilistic classification trees training and test for weakly labelled data. We call "weakly labelled data", training set such as the prior labelling information provided refers to vector that indicates the probabilities for instances to belong to each class. Classification tree typically deals with hard labelled data, in this paper a new procedure is suggested in order to train a tree from weakly labelled data. Considering a random forest, we show how trees can be associated in the test step. An iterative classifier based on the tree is proposed too. The aim is to develop a procedure that modifies the label in order to extract a class. The proposed methods are compared with typical models such as generative and discriminative methods for object recognition and. The considered models are evaluated on standard datasets from UCI and an application to fisheries acoustics is considered.