Unsupervised Selection and Estimation of Finite Mixture Models

We propose a new method for fitting mixture models that performs component selection and does not require external initialization. The novelty of our approach includes: a minimum message length (MML) type model selection criterion; the inclusion of the criterion into the expectation-maximization (EM) algorithm (which also increases its ability to escape from local maxima); an initialization strategy supported on the interpretation of EM as a self-annealing algorithm.