The HPCLUS Procedure

You can find the number of clusters in the data set by specifying NOC=ABC in the PROC HPCLUS statement as follows:

proc hpclus data=sampsio.dmairis maxclusters=9
         NOC=ABC(B=10 minclusters=2 align=PCA criterion=FIRSTPEAK);
   input  sep: pet: ;
   ods output ABCStats=ABCStats1;
run;

These statements create three additional ODS tables, which are shown in Output 5.2.1, Output 5.2.2, and Output 5.2.4. Output 5.2.1 shows the parameters that are used in the aligned box criterion (ABC) method.

Output 5.2.2 shows the statistics that are obtained for each candidate number of clusters.

The following statements plot the gap values:

title "ABC Statistics Plot Using Gap Values";
proc sgplot data= ABCStats1;
   xaxis type=discrete;
   series x= K y= Gap;
run;
title ' ';

Output 5.2.3 shows that the first peak value is at 4. This result is also shown in the estimated number of clusters table, which is shown in Output 5.2.4.

Output 5.2.3: Gap Values

When you use the NOC= option, the HPCLUS procedure first estimates the number of clusters, k, and then it displays the cluster analysis results with k. Output 5.2.5 shows the cluster summary table that is obtained by four clusters.