The Influence of Genetic Initial Algorithm on the Highest Likelihood in Gaussian Mixture Model
Jinjia Wang, Wenxue Hong, Xin Li · 2006
The EM algorithm is a familiar tool to get maximum likelihood parameter estimation in Gaussian mixture model. But the main drawback of EM is that its solution can highly depend on its initial values, and consequently produce sub-optical maximum likelihood estimates. Thus a genetic initialization algorithm (GIA) is proposed to overcome this limitation. K-mean, FCM, and GIA is compared based on several experiments on synthetic and real data sets. Analysis of the experimental results shows that the proposed GIA achieve the highest likelihood