1
$\begingroup$

I have some data set that look like that https://postimg.cc/3y2jx5xR

I am looking for an unsupervised method that can see also the points that start to look different from the majority. Which clustering techniques (I use python) can be used for such data sets?

I have tried k-means but as I was expecting it has failed considerably to see such peaks.

$\endgroup$
2
  • $\begingroup$ Have not tried it myself, but this python package has just released github.com/beringresearch/ivis, for both supervised and unsupervised! $\endgroup$ Jul 30 '19 at 21:13
  • $\begingroup$ I would not consider that to be a cluster. The values are very much spread out. Maybe you should consider anomaly detection instead? $\endgroup$ Jul 31 '19 at 6:32
0
$\begingroup$

You can try with DBSCAN. It's a clustering algorithm that is meant to isolate outliers. You can read more about it here. It's already available in sklearn (documentation here).

If you tweak its parameters correctly, those "extreme" observations should be identified and separated from actual clusters.

$\endgroup$

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.