MATLAB Answers

question about determining the correct number of clusters (documentation)

2 views (last 30 days)
cgo
cgo on 19 Aug 2018
Answered: Bernhard Suhm on 3 Oct 2018
I have a question about determining the correct number of clusters (in k-means clustering).
In the documentation, there is a section about 'determining the correct number of clusters'. Please help me understand what the arguments in the table are for: iter (which would mean iterations), phase (?), num (?), and sum(?).
In the example, you will see:
Best total sum of distances = 1771.1
Here are my questions:
  1. Are we going for the best total sum of distances?
  2. From here, how are we able to determine that 4 is indeed the correct?

  0 Comments

Sign in to comment.

Answers (1)

Bernhard Suhm
Bernhard Suhm on 3 Oct 2018
You can use the silhouette plot or evalclusters function to evaluate the quality of your clustering. There's a little more info in this previous answer.

  0 Comments

Sign in to comment.

Sign in to answer this question.