Gaussian parsimonious clustering models

Abstract : Gaussian clustering models are useful both for understanding and suggesting powerful criteria. Banfield and Raftery (1993) have considered a parametrization of the variance matrix Ek of a cluster Pk in terms of its eigenvalue decomposition, Ek = lkDkAkD'k, lk where lk defines the volume of Pk, Dk is an orthogonal matrix which defines its orientation and Ak is a diagonal matrix with determinant 1 which defines its shape. This parametrization allows us to propose many general clustering criteria from the simplest one (spherical cluster with equal volumes which leads to the classical k-means criterion) to the most complex one (unknown and different volumes, orientations and shapes for all clusters). Methods of optimization to derive the maximum likelihood estimates as well as the practical usefulness of these models are discussed. We especially analyze the influence of the volumes of clusters. We report Monte-Carlo simulations and an application on stellar data which dramatically illustrated the relevance of allowing clusters to have different volumes.
Type de document :
RR-2028, INRIA. 1993
Liste complète des métadonnées
Contributeur : Rapport de Recherche Inria <>
Soumis le : mercredi 24 mai 2006 - 15:58:37
Dernière modification le : vendredi 16 septembre 2016 - 15:11:13
Document(s) archivé(s) le : mardi 12 avril 2011 - 17:58:04



  • HAL Id : inria-00074643, version 1



Gilles Celeux, Gérard Govaert. Gaussian parsimonious clustering models. RR-2028, INRIA. 1993. 〈inria-00074643〉



Consultations de la notice


Téléchargements de fichiers