Clustering Categorical Variables

[koomalkc]

Hi,
Need some tips about how to do clustering for categorical variables.

Any epaper or thoughts are appreciated.

Thanks,

_kamal

[PGStats]

You could use proc distance and choose a measure appropriate for nominal variables to get a distance matrix. Then use that distance matrix in proc cluster.

PG

[SteveDenham]

I wonder if you could get the output from PROC CORRESP, or the necessary internal datasets, and feed those to PROC CLUSTER.

Steve Denham

[PGStats · Accepted solution]

Proc Corresp calculates coordinates for the levels of two or more categorical variables based on their crosstabulation. If you fed distances derived from those coordinates to Proc Cluster, you could cluster together the levels of two or more categorical variables. I never tried doing that.

PG

[koomalkc]

Thanks guys

[turanbul]

Hello,

I am also having a similar problem. As far as I can see using proc distance first and then apply Proc cluster seems appropriate.

What about in Miner? How can I do this with FASTCLUS and MODECLUS?