number of clusters

[Sriram1]

Hi All,

Just wante to know how can we identify  the number of clusters in fastclus apart from the market and business understanding.

Thanks in Advance.

sriram1

[ballardw]

The OUT= data set will have a variable, Cluster by default naming, indentifying each record's assigned cluster.

One simple way is to run Proc freq on the out= dataset and the cluster variable. That will show the number of clusters and how many records were assigned to each cluster.

[Ksharp]

That is a tough question have not been solved by now yet.
You'd check the sas documentation of Cluster Analyst .
In sas documentation ,one way is to use the kth nearest cluster.

But My proposal is using Primary Component Analysis to get two Primary Component ,
And plot these high dimension data into two dimension scatter point based on these
 two Primary Component , and you can roughly see what number of clusters you should take.