Robust clustering via mixture models with contaminated component densities

Initially, a mixture of multivariate contaminated normal distributions was developed for model-based clustering. In addition to the parameters of the classical normal mixture, this model has, for each cluster, a parameter controlling the proportion of mild outliers and one specifying the degree of contamination. Parsimony is introduced via eigen-decomposition of the component covariance matrices. R software was developed to make the approach publicly available. This model was then extended in several ways to account for different situations that may arise.