In linear discriminant analysis, how is the linear discriminant function determined? Assuming equal variance-covariance matrices, is the linear discriminant function determined from the training data?

  • $\begingroup$ The formula given by @Mimshot is only for 2-class situation. See complete set of formulas to get the discriminants for k-class situation here $\endgroup$
    – ttnphns
    Commented Oct 28, 2013 at 17:23

1 Answer 1


For Linear discriminant analysis the linear discriminant function is just the inner product of a given data point $\vec{x}$ with the vector $\vec{w}$, with the criterion $\vec{x} \cdot \vec{w} > c$. The vector $\vec{w}$ is calculated as:

$$\vec{w} = \Sigma^{-1}(\vec{\mu_0} - \vec{\mu_1})$$

Where $\vec{\mu_n}$ is the vector mean of sample class $n$, that is, the mean of the training data for class $n$.

  • $\begingroup$ So you could not compute a linear discriminant function given just one point? $\endgroup$
    – DonutPhil
    Commented Oct 28, 2013 at 15:12
  • $\begingroup$ @DonutPhil If you have just one point, what is there to discriminate? $\endgroup$
    – Nick Cox
    Commented Oct 28, 2013 at 18:02

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.