# Supervised clustering

I have a subset of the original dataset as follows:

Families GH1 GH2 GH3 GH4 GH5 GH6
sp1      23   21  34  23  12  5
sp2      5    19  17  12  90  76
sp3      54   32  13  19  9   0
sp4      1    3   0   12  0   78
sp5      45   24  2   34  21  4


I selected 20 important variables using random forest algorithm. Now, I would like to cluster the families based on these 20 selected variables. I already know that sp1 and sp2 belong to a particular region. Could that information be used in the training set? Could someone explain how to do a supervised clustering? I would prefer an example in R.

• I'd rather do unsupervised clustering, and then reject clusterings where sp1 and sp2 are not clustered as desired, repeat. If it is hard to find a desired clustering, you need to rethink your approach and data. – Anony-Mousse May 11 '17 at 6:31
• I don't understand why this question was downvoted. This is a valid question. – shadowtalker May 16 '17 at 23:15
• I've just barely started learning about this literature myself, but you're looking for constrained clustering. – shadowtalker May 16 '17 at 23:16