Clustering criteria for discrete data and latent class models

Gilles Celeux , Gérard Govaert (1)

1 SAGEP - Simulation, analyse et gestion des systèmes de production

We show that some well known clustering criteria for discrete data, the information criterion and the c2 criterion, are closely related with the classification maximum likelihood criterion for the latent class model. Emphasis is placed on binary clustering criteria which are analyzed under the maximum likelihood approach for different multivariate Bernoulli mixtures. This alternative form of criteria reveals non-apparent aspects of clustering techniques. All the discussed criteria can be optimized with the alternating optimization algorithm.

Mots clés: binary clustering; L1distance; mixture; latent class models

Domaines: Autre [cs.OH]

Gilles Celeux, Gérard Govaert. Clustering criteria for discrete data and latent class models. [Research Report] RR-1122, INRIA. 1989, pp.9. ⟨inria-00075437⟩