B0631
Title: The inverted Dirichlet through a mode viewpoint with clustering applications
Authors: JT Ferreira - University of Pretoria (South Africa) [presenting]
Andriette Bekker - University of Pretoria (South Africa)
Arno Otto - University of Pretoria (South Africa)
Antonio Punzo - University of Catania (Italy)
Salvatore Daniele Tomarchio - University of Catania (Italy)
Abstract: There has been significant interest in the study of flexible and asymmetric models during the last three decades; with some emphasis on the mode as a more "natural" measure of location than the mean or the median. The practical interpretation of the parameters when they are mode-parameterised is of succinct value when considering finite mixtures in a clustering framework. A mode-parameterized inverted Dirichlet (or Dirichlet type II, multivariate inverted beta distribution) is introduced and studied as a candidate to model multivariate data with positive support, and it is demonstrated how the parameterization simplifies its use in various fields of statistics, namely in nonparametric and robust statistics. The interpretability and impact of this model are illustrated using real data within a clustering framework, to emphasise the value of the mode viewpoint.
