My plan is to do supervised dimension reduction and I am planning to try fisher discriminant analysis. My question is that do I need to test if my features are correlated separately to get rid of redundancy or does the fisher discriminant analysis do this job?
In the algorithm I did not explicitly find anything to get rid of correlated features or I might have missed that.
Thank you.
Collinearity can be a problem:
https://stats.stackexchange.com/questions/132774/collinearity-and-linear-discriminant-analysis
Anyway, I guess that your goal is emphasizes class separability.
Dear Zeynab, Fisher discriminant analysis does not take into account descriptors collinearity, to do so you can apply liear discriminant analysis on principal component scores coming from your original variables
Alessandro Giuliani
Dear Alessandro: Thanks for the suggestion. would you please explain more what you mean by "apply linear discriminant analysis on principal component scores coming from your original variables"?
Take your original variables compute on them PCA this procedure Will give rise to n peincipal components (n being the number of variables) take the most important p components (p
Dear Zeynab Mousavikhamene ,
I agree with Alessandro Giuliani that he explains better the third option in the link that I have provided in my answer to your question.