A comparative study of the K - means algorithm and the normal mixture model for clustering : Univariate case

This paper gives a comparative study of the K-means algorithm and the mixture model (MM) method for clustering normal 7 data. The EM algorithm is used to compute the maximum likelihood estimators (MLEs) of the parameters of the MM model. These parameters include mixing proportions, which may be thought of as the prior probabilities of different clusters; the maximum 9 posterior (Bayes) rule is used for clustering. Hence, asymptotically theMMmethod approaches the Bayes rule for known parameters, which is optimal in terms of minimizing the expected misclassification rate (EMCR). 11 The paper gives a thorough analytic comparison of the two methods for the univariate case under both homoscedasticity and heteroscedasticity. Simulation results are given to compare the two methods for a range of sample sizes. The comparison, which 13 is limited to two clusters, shows that the MM method has substantially lower EMCR particularly when the mixing proportions are unbalanced. The two methods have asymptotically the same EMCR under homoscedasticity (resp., heteroscedasticity) when the 15 mixing proportions of the two clusters are equal (resp., not too unequal), but for small samples the MMmethod sometimes performs slightly worse because of the errors in estimating unknown parameters. 17 © 2007 Elsevier B.V. All rights reserved. MSC: 62H30; 62F10 19