Institut Français des Sciences et Technologies des Transports, de l'Aménagement et des Réseaux

Type de document

COMMUNICATION AVEC ACTES INTERNATIONAL (ACTI)

Langue

anglais

Auteur

BILLOT ; SUCHIER ; LALLICH

Résumé / Abstract

This work deals with the adaptation of the main principles of boosting, an efficient supervised ensemble method, into an unsupervised framework. As these two fields are conceptually distinct, boosting unsupervised learning methods raises a certain number of problems. Some are related to clustering and cluster ensembles, such as the clustering validation, the cluster correspondence and the results aggregation. More specific problems arise from the specific case of boosting. In this paper, a new Unsupervised Boosting-Like Approach (UBLA) is proposed, aiming at responding to these critical issues. The experimentations show significant results which pave the way for a better integration of boosting into clustering. eric.univ-lyon2.fr/