Say clustering was performed. My concern is to understand what characterizes certain cluster. Which variables differ most for a certain cluster? Are there any methods for performing such kind of analysis?

For example, say I have 50 variables and 10 clusters. The desired result would sound like: for a cluster 1 variables 5, 16, 23, 42 and 49 contribute to the biggest difference from other clusters.


You can always assume that the clusters are classes and train a classifier on that.

A decision tree classifier comes to my mind, as it will often produce a "human readable" classification tree.

  • $\begingroup$ I like this approach! Is there any literature on it I can cite? $\endgroup$ Mar 9 '15 at 20:25

An interesting topic to look into is Fuzzy clustering.

The idea behind fuzzy clustering is that a cluster can have elements that are more "typical" than others and therefore have a higher membership within that cluster than other members which are less "typical".

For example, a Robin would have a high value (say 0.9) for its membership in the Birds cluster, while a Penguin would have a low value (say 0.4) for its membership in the Birds cluster.

With 50 data items (and 10 clusters!) to cluster a good inference might be difficult to obtain.

A good book to read on Fuzzy Clustering and Fuzzy Logic is the following Fuzzy Sets and Fuzzy Logic: Theory and Applications by Bo and Klir, although it is a bit dated.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.