Maximum Likelihood Estimation of Gaussian Cluster Weighted Models and Relationships with Mixtures of Regression
Cluster-weighted modeling (CWM) is a mixture approach for modeling the joint probability of a response variable and a set of explanatory variables. The parameters are estimated by means of the expectation-maximization algorithm according to the maximum likelihood approach. Under Gaussian assumptions, we analyse the complete-data likelihood function of cluster weighted models. Further, under suitable hypotheses we show that the maximization of the likelihood function of Gaussian cluster weighted models leads to the same parameter estimates of finite mixtures of regression and finite mixtures of regression with concomitant variables. In this sense, the latter ones can be considered as nested models of Gaussian cluster weighted models.