Reputation: 1
I am trying to find an optimum cluster size using the Cluster Node and CCC criterion.
The Automatic setting (default) configures SAS Enterprise Miner to automatically determine the optimum number of clusters to create using either Ward or Centroid method. However, I have some serious problems with the automatic method, the selection of the "optimum" cluster size, and the reported statistics. The options chosen were Cluster Method=Ward, Prelim Max=50, Min=5, Final Max=50, CCC Cutoff=3.
Note the following Cluster Node result. .
The Output shows three candidates for optimum number of clusters k=6, 10 and 46 with CCC=-104, -80 and 163. The best was selected, i.e. k=46 with CCC=163. At the same time, the Cluster Statistics report states that the resulting CCC was 294, however, the maximum CCC within the available range was only 163 (of k <= 50). Interestingly when the preliminary maximum is increased gradually to 500, the discrepancy between the two reported values of CCC is getting smaller.
The CCC vaules shown in the Output are consistent with the chart and the selection of the optimum k for clustering. The CCC value in the cluster statistics table is a pure fantasy and I cannot see any way of translating it to anything meaningful.
Is there something wrong with my analysis?
P.S. I've read the Cluster Node help.
Upvotes: 0
Views: 1383