Skip to content

K-modes clustering : how to choose the number of clusters?

2 messages · 이아름, David L Carlson

#
Dear all. 
 
I tried to cluster the data with categorical variables with K-modes using klaR packages. 
 
I tried to find the optimal number of clusters by considering the average silhouette width though..  
 
In k-modes, however, the average silhouette width increases, when the the number of clusters increases in my case.
 
So i tried to derive the elbow plot and I got the attached graph. 
It is quite hard to which point is the location of a bend in this plot.. 
In this case, how can I choose the best number of groups? 
 
Can anyone introduce better method that help choose the optimal number of clusters for K-modes? 
 
Thanks!
 
Kind regards
 
Ann. 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Example_ElbowPlot.png
Type: image/png
Size: 7087 bytes
Desc: not available
URL: <https://stat.ethz.ch/pipermail/r-help/attachments/20161213/834676ab/attachment.png>
#
Function NbClust() in package NbClust computes 30 different indices for determining the number of clusters.

-------------------------------------
David L Carlson
Department of Anthropology
Texas A&M University
College Station, TX 77840-4352

-----Original Message-----
From: R-help [mailto:r-help-bounces at r-project.org] On Behalf Of ???
Sent: Tuesday, December 13, 2016 5:08 AM
To: r-help at R-project.org
Subject: [R] K-modes clustering : how to choose the number of clusters?

 
Dear all. 
 
I tried to cluster the data with categorical variables with K-modes using klaR packages. 
 
I tried to find the optimal number of clusters by considering the average silhouette width though..  
 
In k-modes, however, the average silhouette width increases, when the the number of clusters increases in my case.
 
So i tried to derive the elbow plot and I got the attached graph. 
It is quite hard to which point is the location of a bend in this plot.. 
In this case, how can I choose the best number of groups? 
 
Can anyone introduce better method that help choose the optimal number of clusters for K-modes? 
 
Thanks!
 
Kind regards
 
Ann.