0

According to this paper, Canonical Discriminant Analysis (CDA) is basically Principal Component Analysis (PCA) followed by Multiple Discriminant Analysis (MDA). I am assuming that MDA is just Multiclass LDA.

My work uses SciKit-Learn's LDA extensively. According to its description, it is

A classifier with a linear decision boundary, generated by fitting class conditional densities to the data and using Bayes’ rule.

It applies pretty well to cases with multiple classes; can this implementation be also called as MDA?

So to do CDA in Scikit-Learn, should one just apply PCA followed by LDA?

Ébe Isaac
  • 800
  • 1
  • 13
  • 26
  • 1
    In my awareness/understanding (see footnote [here](http://stats.stackexchange.com/a/190821/3277)) CDA is a synonym to (multiclass) LDA. Discriminants in multiclass settings are extracted as canonical variates. Do by MDA you mean multi-class LDA? LDA [is not related to PCA closely](http://stats.stackexchange.com/q/169436/3277) because PCA is unsupervised dim. reduction but LDA is supervised dim. red. LDA is a particular case of canonical correlation analysis (CCA) which difference from PCA is explained in pics [here](http://stats.stackexchange.com/q/14002/3277). – ttnphns Mar 08 '17 at 07:36
  • I wrote this comment without having read the article you link to, so I can't appreciate it anyhow. You might wish to search our site for `LDA PCA`, for more info. – ttnphns Mar 08 '17 at 07:38
  • [Here](http://stats.stackexchange.com/a/71571/3277) is an answer with some important further links in it. Also, search threads with tag [tag:discriminant-analysis]. – ttnphns Mar 08 '17 at 07:47
  • @ttnphns: I agree with the unsupervised and supervised differences. It is just that researchers in the specified domain (gait analysis) prefer the term MDA for all their work. They also often use CDA for gait recognition. I just wanted to be sure what the LDA in SciKit Learn is all about before using the term in my paper. – Ébe Isaac Mar 08 '17 at 07:48
  • @amoeba: Thank you but I'm clear with that too. However, is it the same as MDA and is it synonymous with CDA as stated by ttnphns? – Ébe Isaac Mar 08 '17 at 07:54
  • It might be that different people use terms "CDA" and "MDA" differently. These terms are not standard in contemporary statistical / machine learning literature. In the literature I know, "LDA" is used for LDA, no matter how many classes there are. – amoeba Mar 08 '17 at 08:00

0 Answers0