Sunday, July 27, 2008

SAS Tips 1001 - Tip 10 - PROC ACECLUS provides better results if variables are standardized

PROC ACECLUS provides better results if variables are standardized
See the discussion in
http://support.sas.com/documentation/cdl/en/statug/59654/HTML/default/statug_stdize_sect020.htm#statug.stdize.stdizesummary
Interestingly, the best standardizations with in 5 observations missclassifications (3% of observations missclassified) are (SPACING (0.14)-25, MAXABS-26, IQR, AGK(0.14)-28, RANGE-32, MIDRANGE-32. All these suggest 7 clusters.

Interestingly, STD and L(2) well known looks like the best because both identify only 5 clusters with missclassifications of 33 obs!

No comments: