“…the model with the optimum number of clusters and the best covariance structure) achieves high similarities within each cluster and low inter-similarities between clusters [26,27]. Several studies have been performed on model selection such as the Akaike information criterion (AIC), the Bayesian information criteria (BIC) [28], neural network methods [29,30] and a Bayesian approach that is combined with Monte Carlo integration [31]. The BIC, unlike AIC, considers the dimension of input space by including the number of features as a penalty [28,32], and thus the BIC is more sensible.…”