TITLE:  Gaussian Mixture Estimation With Incomplete Models


ABSTRACT:  A common use of mixture modeling in multivariate
data is for the estimation of clusters.  The practical
difficulty is the determination of the correct number of
components in the mixture model.  This choice is confounded
by the need for reasonable initial guesses for the parameters
for each mixture component.  In this paper, we describe a
hybrid algorithm which attempts to fit incomplete mixture
models, or partial mixture component models.  Bias problems
are avoided since the partial mixture model is fitted many
times using carefully chosen random starting guesses.
Clustering the results of these many fits using an advanced
visualization tool such as GGobi can reveal individual
components.  We describe the algorithms, some simulations,
and some real experiments.