Abstract

Recently, a paper published by Celeux et al. (2006) presented several forms for the deviation information criterion (DIC) for mixture models, each version is depended on the kind of probability function. However, no reliable version was adopted for those models. As an idea inspired by Brooks (2002, p. 617), we develop, in this paper, Bayesian deviations plugging into two known criteria: the Akaike information criterion (AIC) and Bayesian information criterion (BIC) for choosing best mix model. Due to unavailability the closed-form of the perceived likelihood of those models, we propose an algorithm for estimating the observed likelihood for mixture models via an Markov chain Monte Carlo (MCMC) approach. It is shown via recreation researches and examples include actual information applications that proposed AIC and BIC perform well

Keywords: Finite mixture models observed likelihood Gibbs sampler model selection

DOI:10.52113/6/2020-10-2/137-153