To cluster or not to cluster: that is the question

From: Stats Wolf <stats.wolf@xxxxxxxxx>
Date: Fri, 11 Apr 2008 11:12:57 -0700 (PDT)

Hi all,

I am using k-means algorithm, and to detect the optimum number of
clusters, Calinski and Harabasz's C(g) coefficient. Unfortunately,
their approach does not enable me to determine the C(g) for one
cluster (so, for the situation when I actually don't cluster the
sample). If the optimum number of clusters is higher than 2, that's
fine, but when it is actually 2, I can't know whether 2 is indeed the
optimum number of clusters or maybe I should not cluster the sample
whatsoever. What do you think about this?

Thanks in advance,
Wolf
.

Follow-Ups:
- Re: To cluster or not to cluster: that is the question
  - From: Ray Koopman
- Re: To cluster or not to cluster: that is the question
  - From: Art Kendall

Prev by Date: Re: Specific comparisons in binomial GLM
Next by Date: Bayesian Regression - goodness of fit
Previous by thread: Q: dichotomous variables
Next by thread: Re: To cluster or not to cluster: that is the question
Index(es):
- Date
- Thread

Relevant Pages

Re: To cluster or not to cluster: that is the question
... Calinski and Harabasz's Ccoefficient. ... If the optimum number of clusters is higher than 2, ...
(sci.stat.consult)
Re: To cluster or not to cluster: that is the question
... K-means can give you different solutions depending on the order cases are sorted in. ... The classification society deals specifically with clustering, multidimensional scaling, etc. ... To help you decide on the number of clusters to retain, it gives AIC or BIC for varying numbers of clusters. ... If the optimum number of clusters is higher than 2, ...
(sci.stat.consult)