Take a look at my original data. (masked with purely random alphabetic here) :

    a b c d e
    f g h i j
A = k l m n o
    p q r s t
    u v w x y

I'm running kMeans (3 cluster) on the data, resulting final centroid like this :

    aa aa aa aa aa
B = bb bb bb bb bb
    cc cc cc cc cc

Now, before run kMean again, I applied PCA on data and took only first three principal component :

    xa yb zc
    xd ye zf
C = xg yh zi
    xj yk zl
    xm yn zo

After that, I ran kMeans for 3 centroid and, of course, resulting 3 centroid :

    xaa yaa zcc
D = xbb ybb zbb
    xcc ycc zcc

The cluster result with PCA are exactly same with the first test (without PCA). My question : After finishing kMeans with PCA, can I say that this cluster (say cluster 1) has centroid aa aa aa aa aa, rather than saying that cluster 1 has centroid xaa yaa zcc?

  • $\begingroup$ Since both cluster analyses in your specific case gave the same results - the same distribution of cases among clusters - then yes you may of course say it, it's obvious. $\endgroup$
    – ttnphns
    Apr 22, 2016 at 7:01

1 Answer 1


By definition of a linear transformation (and PCA yields a linear transformation),

$$ M\cdot \mu = M\cdot (\frac{1}{N}\sum_i \vec{x}_i) = \frac{1}{N}\sum_i M\cdot\vec{x}_i $$

I.e. if objects are assigned to the same clusters, then the projection of the original centroid must be the same as the centroid of the projections.

That means PCA did not change anything. Did it work?


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.