Aspectos práticos da estimação do modelo de mistura via processo de Dirichlet

Abstract

We review the Dirichlet process mixture model and investigate its performance as a classification method. The first aspect considered is its sensibility to the choice of location parameter of the base distribution. The second aspect considers the performance of the model regarding the departure of the parameters of the component distributions. Simulation results with mixture of normal distributions indicate sensibility to location parameters choices, of the base distribution, and good performance even when components with normal distributions differ only in variances. Finally, we apply the method to three data sets.