Abstract
Deciding on the number of “classes” has been the most prominent and most debated challenge in finite mixture modeling. Recently, a novel strategy has been proposed to select the best model in finite mixture modeling: a k-fold cross-validation approach. However, this approach has not been systematically evaluated, which makes the performance of the k-fold cross-validation approach for model selection in finite mixture modeling largely unknown. Thus, the main motivation for conducting the current work is to systematically evaluate the performance of the k-fold cross-validation approach for model selection in the context of Growth Mixture Modeling. Results revealed that the performance of the k-fold cross-validation approach for model selection in GMM is generally unsatisfactory, and it only performs reasonably well under the condition of very large class separation.