1CLUSTERING MULTINOMIAL OBSERVATIONS VIA FINITE AND INFINITE MIXTURE MODELS AND MCMC ALGORITHMS

Traditional statistical clustering procedures based on finite mixtures model require the number of mixture components to be known prior to the analysis. Establishing the number of mixture components from the data is generally a difficult problem involving the comparison of models with different number of parameters. We used the Bayesian infinite mixture approach to devise a clustering procedure that does not require the number of mixture components to be specified in advance. The performance of this model is compared to the performance of the finite mixture approach when the number of components is known as well as when the number of components is estimated using AIC criterion. We showed that the infinite mixture procedure offers comparable results to the finite mixtures approach