ORDER ESTIMATION FOR MULTIVARIATE MIXTURE REGRESSION MODELS

Abstract In this paper, we study the problem of jointly selecting the number of components and explanatory variables for multivariate mixture regression models. In practice, the selection model using the Akaike Information Criterion AIC is not satisfactory, as it can lead to an overestimation of the number of components and the number of explanatory variables. To improve selection, Naik et al. (2007) developed a new criterion based on Akaike’s technique, the Mixture Regression Criterion MRC for the simultaneous determination of the number of components and explanatory variables for univariate mixture regression models. We propose a generalization of the criterion MRC for multivariate mixture regression models. The performance of the new criterion is validated on simulated data by comparing it to the Akaike criterion AIC and the Schwarz criterion BIC.