Im trying different linkage methods for my hierarchical clustering problem. Now I would like to evaluate which one works better. Is this as easy as just just comparing the two Dunn's index values? Or should I take other factors into consideration?
$\begingroup$
$\endgroup$
3
-
$\begingroup$ Which linkage methods are you using? This is valuable information when it comes to suggestions for comparison methods. $\endgroup$– AndyCommented Jun 1, 2016 at 14:56
-
$\begingroup$ Im using single - and complete linkage $\endgroup$– Frits VerstratenCommented Jun 1, 2016 at 14:57
-
5$\begingroup$ Does this answer your question? How to select a clustering method? How to validate a cluster solution (to warrant the method choice)? $\endgroup$– ttnphnsCommented Apr 21, 2020 at 21:22
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Dunn's index is the ratio of minimum inter-cluster distance to maximum cluster diameter found. As such it is a good measure to compare the two methods. The one thing to look out for with Dunn's index is if one cluster is particularly large while all others are compact, then the metric will be misleadingly low.
Other metrics to consider could be:
within clusters sum of squares
between clusters sum of squares
