What's the difference between outlier analysis and clustering?

When clustering is done isn't that outliers (if exist) are also found?


1 Answer 1


No, clustering creates a pre-defined number of clusters $k$, so all "outliers" will end up in in one of these "common" clusters. If you increase $k$ you may eventually be able to form some clusters, which are far off from the other clusters, and hence find some "outliers", however there is no guarantee for this. Also if you have only a small amount of "outliers" then the algorithm probably won't cluster these in their own clusters but rather join them in some other bigger cluster.

Probably the best option would be hierarchical clustering with single linkage, as this would create a hierarchical structure based on distance, from where you could find some very distant branches based on plots, but still this is not what it was intended for.

  • $\begingroup$ Not all clustering methods require the number of clusters to be predetermined. $\endgroup$
    – whuber
    Commented Nov 25, 2019 at 16:58
  • $\begingroup$ @whuber Are you referring to methods where you select the number of clusters based on some criterion? $\endgroup$ Commented Nov 26, 2019 at 8:08
  • $\begingroup$ Yes--and that would include methods to identify an arbitrary number of outliers. $\endgroup$
    – whuber
    Commented Nov 26, 2019 at 13:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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