Analiza skupień oparta na mieszankach uciętych rozkładów normalnych
Trimming approach to the mixtures of normal distributions

Summary/Abstract: The paper presents trimming approach to the mixtures of normal distributions. In this approach the proportion of the most outlying observations is trimmed (during parameter estimation), different constraints on the cluster scatter matrices are assumed and trimming likelihood is defined. To estimate the parameter of this kind of mixture mostly the modified version of EM, i.e. TCLUST algorithm is applied [García-Escudero et al. 2010, p. 89-109]. The results of clustering given by trimming approach and well known robust mixtures of normal distributions are compared.

• Details
• Contents