Comparison of Estimation Techniques for Mixtures of Normal Densities
Department of Statistics
Colorado State University
Monday, 19 May 2003
E103 Engineering Building
This presentation introduces a new method that can improve the performance
of existing techniques for estimating parameters in mixtures of
normal densities. A simulation study compares the new method
to a Bayesian method of mixture density estimation and four non-Bayesian
methods. This study is most likely the first simulation testing
of the novel technique and the Bayesian technique. A thorough
review of the literature did not discover any previous testing.
In addition to simulations, the performance of the six methods is
compared in an application to three real data sets. The conclusions
of this study offer recommendations regarding which method could
best handle mixture data with specific characteristics.