The paper introduces a methodology for visualizing on a dimension reducedsubspace the classification structure and the geometric characteristics inducedby an estimated Gaussian mixture model for discriminant analysis. Inparticular, we consider the case of mixture of mixture models with varyingparametrization which allow for parsimonious models. The approach is anextension of an existing work on reducing dimensionality for model-basedclustering based on Gaussian mixtures. Information on the dimension reductionsubspace is provided by the variation on class locations and, depending on theestimated mixture model, on the variation on class dispersions. Projectionsalong the estimated directions provide summary plots which help to visualizethe structure of the classes and their characteristics. A suitable modificationof the method allows us to recover the most discriminant directions, i.e.,those that show maximal separation among classes. The approach is illustratedusing simulated and real data.
展开▼
