"Newbie: Cluster based on multiple attributes?"

IngoRM

Original message from SourceForge forum at http://sourceforge.net/forum/forum.php?thread_id=2048818&;forum_id=390413


I am looking to cluster some 2D points based on their proximity to each other. I was looking at using the AgglomerativeClustering operator, but haven't been able to figure out what parameters to use.

Any suggestions on how to do this would be much appreciated.

IngoRM

Hi,

basically all cluster schemes try to group similar points while not so similar points should be part of other groups. If you have points in a 2D plane you are very lucky because you could visualize the clustering and check for cluster validity yourself instread of relying on some cluster quality measurements.

Personally, I would start with schemes like k-means or DBScan before using a hierarchical clusterer in this case. About the parameters for the latter: what is you concrete question? In general, there is no silver bullet and you probably will have to test different parameter combinations in order to get good results. Then again it is nice that you only have two dimensions...

Cheers,
Ingo