I have samplings of one-dimensional data of two classes: $A$ and $B$.

I have to predict the posterior probability of class $A$.

$$\tilde{P}(A|x) \approx \frac{N_A\tilde{f_A}(x)}{N_A\tilde{f_A}(x) + N_B\tilde{f_B}(x)}.$$

$f_A$, $f_B$ are the estimation of densities of classes $A$ and $B$.

I estimate the densities as normal distributions, whose $\mu$ and $\sigma$ are average values and standard deviations of all the points of each class ($A$ and $B$ accordingly).

Will it be correct to say, that I'm using discriminant analysis in this case?

  • $\begingroup$ I'm not fluent in deciphering formulae, but what I may say for sure is that discriminant analysis (see what's written in its tag) first extracts latent dimensions (discriminants), like PCA does. Only then it classifies - with those. As far as I can see, your question is about classification only. It is unclear if you are speaking of the original variables or the discriminant latent variables here. $\endgroup$ – ttnphns Oct 15 '13 at 14:47
  • $\begingroup$ There is only one dimension. What discriminants can we extract here? $\endgroup$ – Felix Oct 15 '13 at 15:38
  • $\begingroup$ Ah, well (I didn't see it). Then term "discriminant analysis" proper is inapplicable, although word "discrimination" can retained, as it is close to "classification" or "distinguishing". $\endgroup$ – ttnphns Oct 15 '13 at 19:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.