I've 110 variables & 200 data points. Of this 110 variables, one is group variable (say "brown eye","blue eye"). I want to use discriminant analysis to classify the groups based on remaining 119 variables. Since the variables are large, to get a meaningful result I need to reduce the number of variables. So, the are 3 options to me:

1) Stepwise Discriminant Analysis: I don't want to use this method as I'm biased towards it.

2) Classification Tree Method: This method will give an idea about which variables affecting the eye color. Since the dataset is small, I'm apprehensive of using this method.

3) Principal Component Method: This method I can use. But I prefer to keep the original variables.

My question is can anybody please suggest me some other method to select variables for discriminant analysis.

  • $\begingroup$ Wonder about (1). Could you comment on your bias? As for suggestion, one possible method is stepwise nominal logistic regression. But you have too many variables to use it, for me. $\endgroup$
    – ttnphns
    Aug 13, 2011 at 7:05
  • $\begingroup$ I don't like stepwise as it may have include some unimportant variables. $\endgroup$
    – Beta
    Aug 13, 2011 at 7:59
  • 1
    $\begingroup$ Hmm ? Well. What is "important" for you? Points (1) and (2) imply important as predictor. Point (3) imply important as a descriptor of multivariate cloud. These are different concepts. Have you made up your mind on 'importantness"? $\endgroup$
    – ttnphns
    Aug 13, 2011 at 8:41

1 Answer 1


You can get rid of some by looking for pairs that are very highly correlated and randomly deleting one of the pair.

Then you can look at partial least squares, and pick variables that are important in the PLS solution.

I did this with a similar problem and it worked pretty well (that is, the resulting discriminant function did pretty well)

  • 1
    $\begingroup$ Thanks Peter for your answer.When nobody was answering, I was started thinking that I asked a wrong question :).I was also wondering if LASSO or LARS method can be used in discriminant analysis. I read about them as good variable choosing method in the book "The elements of statistical learning: data mining, inference, and prediction", but don't know how to use it in my present framework. $\endgroup$
    – Beta
    Aug 15, 2011 at 16:07

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.