Variable Selection in Clustering by Mixture Models for Discrete
Data // An implementation of a variable selection procedure in
clustering by mixture models for discrete data (clustMMDD).
Genotype data are examples of such data with two unordered
observations (alleles) at each locus for diploid individual.
The two-fold problem of variable selection and clustering is
seen as a model selection problem where competing models are
characterized by the number of clusters K, and the subset S of
clustering variables. Competing models are compared by
penalized maximum likelihood criteria. We considered asymptotic
criteria such as Akaike and Bayesian Information criteria, and
a family of penalized criteria with penalty function to be data
driven calibrated.