A new approach of genetic-based EM algorithm for mixture models

Date

2011-05

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Finite mixture models have been receiving an important attention over the years in practical and theoretical point of view, but it is still challenging task to estimate a reasonable estimator based on the maximum likelihood method. The most widely used technique to solve the problem, to some extent, is the EM algorithm. Researchers have done a lot of work to improve the results of the EM algorithm by modifying its basic idea. This work present such an attempt to obtain better estimates for a finite normal mixture model. A traditional evolutionary technique, known as Genetic algorithm, is coupled with the EM algorithm to improve the estimates of the EM algorithm started with a random initial vector of parameters. The presented method is tested with the availability of a Non-penalized and Penalized likelihood functions. Based on results, we can see that the proposed method is always superior to the classical EM algorithm when one concerns the global maximizer in the mixture likelihood function.

Description

Keywords

Finite mixture models, Genetic algorithms

Citation