On June 19th 2015,  11.30h at Dipartimento di Statistica e Metodi Quantitativi 

Sala del Consiglio (edificio U7, 4° piano, 4064) Via Bicocca degli Arcimboldi, 8 – 20126 Milano

“Robust model-based clustering and mixture modeling by using trimming and restrictions”

Abstract

The use of trimming procedures is common in many statistical settings to achieve 

robustness against outlying observations. Notably, to obtain robustness in model-based 

clustering and mixture modeling, the only adoption of trimming is not enough. Constraints 

on the scatters, corresponding to the clusters or to the mixture components, are needed 

too. The talk presents robust methodology based on the joint application of trimming and 

feasible restrictions for different clusters and mixture models settings. As it is typical in all 

proposals designed for this aim, some input parameter needs to chosen, to set the 

trimming level and the scatter constraints. Data driven tools for helping the user in 

choosing such values will be also presented.

for more info:

e-mail: francesca.greselin@unimib.it   http://www.dismeq.unimib.it