How to calculate Davies-Bouldin Index in SAS for K-Means

[manojkuchi]

#1

How Does SAS calculate Dunn Index or Davies-Bouldin Index for K-means clustering?

Kindly help.

[sbxkoenk]

Hello,

I have done a lot of k-means clustering in the recent past and I have to say that, unfortunately, the Davies–Bouldin index (DBI) and Dunn's separation index are not standardly available in proc FastClus (SAS/Stat) nor in proc kClus (Visual Statistics). Nor is the Silhouette score (another technique often used for evaluating clustering algorithms and/or determining the optimal number of clusters). SAS uses many typical statistical measures like RMSSTD (root-mean-square standard deviation) and the like for these purposes.

For DBI, you will need to start coding. You can probably use the distance measure (variable) output by the k-means clustering algo (distance of observation to the cluster-center as well as the cluster assignment / cluster membership are available in the out= data set).

There is also a useful procedure called PROC DISTANCE (SAS/Stat). Just mentioning it in case you would need it.

Good luck,

Koen

[sbxkoenk]

In addition to my previous response, ... this paper may also be of interest to you:

Paper 3409-2019 (SAS Global Forum 2019)
How to Evaluate Different Clustering Results
Ralph Abbey, SAS Institute Inc.

https://www.sas.com/content/dam/SAS/support/en/sas-global-forum-proceedings/2019/3409-2019.pdf

Cheers,

Koen