In this work we present an exhaustive simulation study aimed at evaluating the clustering performance of a novel dimensionally reduced model based clustering method, named Factor Mixture Analysis, extended to allow for covariate effects. Factor Mixture Analysis is a particular factor model in which the traditional assumption of Gaussian distributed factors is replaced by a finite mixture of multivariate Gaussians. Since modeling the factors as a multivariate Gaussian mixture amounts to model the observed variable as a particular multivariate Gaussian mixture too, the proposed approach performs clustering and dimension reduction simultaneously. It is further assumed that within each mixture component the factor scores are linearly affected by covariates which may act in a di®erent way as the groups vary.
A simulation study for evaluating the clustering performance of Factor Mixture Analysis with covariates
MONTANARI, ANGELA;VIROLI, CINZIA
2009
Abstract
In this work we present an exhaustive simulation study aimed at evaluating the clustering performance of a novel dimensionally reduced model based clustering method, named Factor Mixture Analysis, extended to allow for covariate effects. Factor Mixture Analysis is a particular factor model in which the traditional assumption of Gaussian distributed factors is replaced by a finite mixture of multivariate Gaussians. Since modeling the factors as a multivariate Gaussian mixture amounts to model the observed variable as a particular multivariate Gaussian mixture too, the proposed approach performs clustering and dimension reduction simultaneously. It is further assumed that within each mixture component the factor scores are linearly affected by covariates which may act in a di®erent way as the groups vary.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
