On classifying mixture components as clusters
N. T. Longford and J. Bartosova
Abstract
An index is defined for characterising the separation of two distributions
and is applied to assessing whether a mixture component is a cluster.
The properties of the index are explored on examples and by simulations.
For a mixture we define a sequence of clustering structures
for a continuum of thresholds that qualify a cluster.
The approach is equally suitable for univariate and multivariate outcomes
with arbitrary distributions of the mixture components.
Submitted; October 2011