2
$\begingroup$

Here is the basic SKLearn tutorial on K-means.

They run PCA and then do K-means on reduced data. Can it radically affect the result? Will we get totally different clusters if we apply PCA on already clustered data?

Improve this question
$\endgroup$
2
  • 3
    $\begingroup$ The topic "k means after PCA" is like "is there life after love" - asked and answered numerous times on this site. Please try to search and read, before asking. $\endgroup$
    – ttnphns
    Commented Nov 21, 2017 at 9:34
  • $\begingroup$ See this answer which in summary says: PCA mitigates the effect of outliers and hence in the presence of outliers, the results could be radically different. stats.stackexchange.com/a/12856/73547 $\endgroup$
    – discipulus
    Commented Feb 2, 2019 at 4:51

1 Answer 1

Reset to default
2
$\begingroup$

Yes, it can radically affect the result.

You can get completely orthogonal clusters if the variances of the components are different by a large factor.

Improve this answer
$\endgroup$

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.