Τρίτη, 03 Δεκεμβρίου 2019, Αίθουσα Μ2, 13:30-14:30
Ομιλητής: Δημήτριος Καρλής, Οικονομικό Πανεπιστήμιο Αθηνών
Τίτλος: Model Based Clustering through Copulas for High Dimensional Data
Περίληψη: In a recent paper Kosmidis and Karlis (2016) proposed model based clustering based on multivariate distributions defined through copulas. This approach offers a number of advantages over existing methods mainly due to the flexibility to define appropriate models in certain different circumstances. In this talk we exploit the ideas of extending the approach for higher dimensions. The central idea is to use a Gaussian copula and implement the correlation matrix of the Gaussian copula through certain parsimonious representations giving rise to models of different complexity. We use two different approaches, the first makes use of factor analyzers based on the factor decomposition of the correlation matrix and the second is based on Choleski type decompositions. Application with real and simulated data will be also described.