Clustering Analysis

bbarnali — Fri, 24 Nov 2017 19:35:38 GMT

Hi All, do you know how to calculate Hopkin Statistic in SAS? Hopkin Statistic tells us whether the dataset has a natural cluster tendency. Thanks in advance 🙂

Re: Clustering Analysis

DougWielenga — Wed, 29 Nov 2017 20:31:02 GMT

Unfortunately, the Hopkins statistic is not currently produced by any procedure in SAS.

PROC MODECLUS performs the saddle test and you can use this to test for unimodality. You need to specify the R= and JOIN options in the MODECLUS statement. METHOD=1 is the most commonly used method and you can find the description here under PROC MODECLUS > Clustering Methods

There is an example in the documentation: Cluster Analysis with Significance Tests

One way to estimate the number of clusters is to run the analysis over a range of values of the smoothing parameter. If you get the same number of clusters over a range of values, then that is a likely indication of the number of clusters. For a more detailed discussion, take a look in the Details section of the MODECLUS chapter, particularly the 'Density Estimation' and 'Significance Tests' sub sections.

For additional discussion and ideas for determining the number of clusters, see "The Number of Clusters".

Hope this helps!

Doug

topic Re: Clustering Analysis in SAS Data Science

Clustering Analysis

Re: Clustering Analysis