Families of Parsimonious Finite Mixtures of Regression Models

Model-based clustering has become increasingly popular during the last decade. Parametric mixture models are used in model-based clustering; however, such models generally do not exploit covariates. Incorporating a regression structure can yield important insight when there is a regression relationship between some variables. Methodologies that deal with such data include finite mixtures of regressions (FMR; [7, 13]) and finite mixtures of regressions with concomitant variables (FMRC; [22]), supported by the popular flexmix package [13]. Multivariate correlated responses can be naturally integrated into such models. However, flexmix currently does not account for correlated response variables for both FMR and FMRC. FMR models that deal with correlated response variables have recently been proposed [19, 9].