New proposals for clustering based on trimming and restrictions

TCLUST is a model-based clustering methodology, which employs trimming and restrictions for getting robust estimators. It is available in the tclust package at the CRAN website and in the FSDA Matlab library. Extensions of TCLUST modelling include clustering around linear subspaces, factor analyzers approaches and fuzzy proposals. Further research has been focused in allowing more flexible models for the components, based on the skew normal distribution. An important issue that may appear within TCLUST is the dependence of the obtained solutions from the input parameters. Therefore, a variety of tools have been developed to assist to the users in choosing these parameters. Theoretical and robustness properties for the TCLUST estimators have been proven, and many empirical evidences show the efficacy of the proposed methodology, in a wide variety of situations