RT info:eu-repo/semantics/preprint
T1 Finding the Number of Groups in Model-Based Clustering via Constrained Likelihoods
A1 Cerioli, Andrea
A1 García Escudero, Luis Ángel
A1 Mayo Iscar, Agustín
A1 Riani, Marco
K1 Estadística
AB Deciding the number of clusters k is one of the most difficult problems in ClusterAnalysis. For this purpose, complexity-penalized likelihood approaches have beenintroduced in model-based clustering, such as the well known BIC and ICL criteria.However, the classification/mixture likelihoods considered in these approachesare unbounded without any constraint on the cluster scatter matrices. Constraintsalso prevent traditional EM and CEM algorithms from being trapped in (spurious)local maxima. Controlling the maximal ratio between the eigenvalues of the scattermatrices to be smaller than a fixed constant c ≥ 1 is a sensible idea for setting suchconstraints. A new penalized likelihood criterion which takes into account the highermodel complexity that a higher value of c entails, is proposed. Based on this criterion,a novel and fully automatized procedure, leading to a small ranked list of optimal(k; c) couples is provided. Its performance is assessed both in empirical examples andthrough a simulation study as a function of cluster overlap.
YR 2016
FD 2016
LK http://uvadoc.uva.es/handle/10324/18093
UL http://uvadoc.uva.es/handle/10324/18093
LA spa
DS UVaDOC
RD 18-may-2024