Gaussian finite-mixture models are extended to include the use of auxiliary information, the dependence of component membership probabilities being modelled by a generalized linear model for polytomous responses. Among the possible applications of the proposed methodology are probabilistic classification and estimation of group conditional parameters. Identifiability features of such a model are investigated in comparison with standard finite mixtures. A full Bayesian hierarchical representation of the model is developed to implement the Gibbs sampling estimation algorithm. Two examples are presented where the methodology is applied to the analysis of real and synthetic data.
A hierarchical finite mixture model for Bayesian classification in the presence of auxiliary information
POLLICE, Alessio;BILANCIA, Massimo
2000-01-01
Abstract
Gaussian finite-mixture models are extended to include the use of auxiliary information, the dependence of component membership probabilities being modelled by a generalized linear model for polytomous responses. Among the possible applications of the proposed methodology are probabilistic classification and estimation of group conditional parameters. Identifiability features of such a model are investigated in comparison with standard finite mixtures. A full Bayesian hierarchical representation of the model is developed to implement the Gibbs sampling estimation algorithm. Two examples are presented where the methodology is applied to the analysis of real and synthetic data.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
