Re: Cluster Analysis - K means or hierarchical?
Posted by
Jon K Peck on
Jun 20, 2012; 1:42pm
URL: http://spssx-discussion.165.s1.nabble.com/Cluster-Analysis-K-means-or-hierarchical-tp5713716p5713720.html
Try using as the k-means starting values
the results from the hierarchical clustering. K-means is sensitive
to starting values. Reordering the data randomly may also help.
Jon Peck (no "h") aka Kim
Senior Software Engineer, IBM
[hidden email]
new phone: 720-342-5621
From:
Shailaja Rego <[hidden email]>
To:
[hidden email]
Date:
06/20/2012 07:36 AM
Subject:
[SPSSX-L] Cluster
Analysis - K means or hierarchical?
Sent by:
"SPSSX(r)
Discussion" <[hidden email]>
I need help in my research - I have 24 variables and
290 valid cases for
cluster analysis. When I perform hierarchical clustering,(Walds method)
it
gives me almost perfect three clusters (85, 90 and 115 cases in each
clusters) the ANOVA is rejected for each of these 24 variables. I have
also
performed Discriminant analysis which gave me 91.5% accuracy in original
grouped cases and 90.5% in cross validated (leave one out ) cases.
My problem is, when I am trying to do a K-means cluster analysis for 3
clusters, Its not giving me proper results, ( 286 in one cluster and 2
each
in other)
Even a two cluster solution for k-means is not proper (2 in one cluster
and
others in other)
Can anyone tell me what's the problem?
I am doing this for my PhD thesis and my guide is insisting on using k-means
cluster saying its more robust. Is this true? Is it not possible
for me to
use hierarchical cluster analysis?
Please help me.
Thanks in advance
Shailaja
--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Cluster-Analysis-K-means-or-hierarchical-tp5713716.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.
=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD