10
$\begingroup$

In PCA eigenvalues determine the order of components. In ICA I am using kurtosis to obtain the ordering. What are some accepted methods to assess the number, (given I have the order) of components that are singificant apart from prior knowledge about the signal?

$\endgroup$
1
  • 1
    $\begingroup$ I actually think that in ICA you can still use the number of 'significant' (ie, 90% of energy) eigen-vectors, as the number of independent components. $\endgroup$
    – Spacey
    Aug 29, 2012 at 15:27

1 Answer 1

1
$\begingroup$

Seeing this question still lacks an answer, I'd like to reiterate, as @Tarantula commented, an accepted method to select assess the number of components retained is through the explained variance. You retain $K$ components given a criterion from the a priori whitening PCA and perform ICA on those components.

I don't know of any accepted method to do this kind of assessment with the kurtosis, this question might be unanswerable in it's fully.

$\endgroup$

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.