Comparing approaches for testing the number of components in a finite mixture model (Q1965946)

The results of numerical experiments for comparing five different procedures for testing the number \(k\) of components of finite mixture models are presented. The following approaches are considered: Orlov's nested tests, Wolf's approach, Aitkin-Rubin's approach, SEM algorithm and McLachan's Monte-Carlo test. The author considers the case of a mixture model of two Gaussian components and tested \(k=1\) against \(k=2\).