K-Means Best K value

ssk1974

Hi,

I read few posts on selecting the best K value for K-means. I manually ran K-Means with few different K Values and a performance operator with Davies Bouldin. These are the values, and I am not sure how to interpret it. Please help.

K-Value Davies Bouldin
2 -0.664
3 -1.017
4 -1.039
5 -0.917
6 -0.881
7 -0.001
8 -0.831
9 -0.855
10 -0.819

1) Which is the optimal K-Value in the above, how do I find it
2) I did this manually, how can I do it using a loop parameter.

Thank you for your time.

awchisholm

Hello

For Davies-Bouldin the minimum shows the most interesting clustering. In general see my blog for many examples of cluster validity measures and how to automate applying them to clusters.

http://rapidminernotes.blogspot.co.uk/search/label/ClusterValidity

In particular see http://rapidminernotes.blogspot.co.uk/2011/03/counting-clusters-part-ii.html

regards

Andrew

ssk1974

Thank you Andrew. In the list I have for Davies Bouldin value, which one is optimal, as they all are negative values, I am not sure which would e considered as minimal.

awchisholm

The minimum - i.e. -1.039.

ssk1974

awchisholm wrote:
The minimum - i.e. -1.039.

Thank you Andrew.