2
$\begingroup$

I have data that I'm trying to classify into two different groups using Fisher linear discriminant analysis. This gives me a vector of weights $\vec w$, used in the equation $\vec w\cdot \vec x$ to give a value that's compared to a threshold in the classification stage.

What I'd like to know is, what kind of information can I extract from this vector $\vec w$ (from the magnitude of weights, their sign, etc.)? Can this, for example, tell me how much information a certain dimension gives about the class?

$\endgroup$
  • 2
    $\begingroup$ LDA extracts the discriminant function - a latent variable that best discriminates your two groups. This function is a linear combination of the input variables. The coefficients in this linear combination are the weights you are speaking about. A weight tells about the discriminative ability of a variable; it tells nothing about a class. $\endgroup$ – ttnphns Aug 2 '13 at 19:09
0
$\begingroup$

you can treat w as a direction. you project x onto that direction (like you decompose a 3-dim vector into 3 components X, Y and Z), and w.x is then the component of x in w direction. because the vectors in different classes are supposed to have different components in that direction, so you can use it for classification purposes.

$\endgroup$

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.