2

I have a set of data which contains a mix of continuous, categorical variables etc. I wanted to apply linear discriminant analysis however, from some research on the Internet I understand that categorical independent variables are not necessarily suited for LDA.

I just wanted to ask, is there a way in which I can alter certain variables so that they are suitable for LDA? Or my question may be, in fact, do I actually need to do this?

Thanks

Akash
  • 21
  • 1
  • 2
  • If you are trying to discriminate among categories using a multivariate normal, LDA will work fine. You cannot use LDA to discriminate among categories using a mix of continuous & categorical variables. You will have to use something other than LDA. – gung - Reinstate Monica Jan 06 '15 at 07:28
  • @gung My data is related to credit scoring where I have the dependent variable (good or bad customer) and then a bunch of explanatory variables such as age, number of instalments, purpose for loan, employment status etc). So if I wanted to carry out LDA on this, are you saying it isn't possible? I wanted to use LDA on this example and then in the evaluation aspect of the report I am writing, state some of the underlying assumptions of LDA and how if they aren't fulfilled then it won't be the best method to use. – Akash Jan 06 '15 at 07:35
  • 1
    Linear discriminant analysis maths does not allow for categorical explanatory variables. Moreover, Bayes' classification of cases within LDA assumes that the variables come from multivariate normal population. However, one may recall that LDA is, in a sense, [a case of](http://stats.stackexchange.com/a/31468/3277) Canonical correlation analysis (CCA). – ttnphns Jan 06 '15 at 08:22
  • 1
    (cont.) But there exist procedure called OVERALS, categorical CCA with optimal scaling approach which can transform categorical variables into quantitative "optimaly" for the task. So, you can perform discriminant analysis by it; you can also use the quantified variables output by it to input into regular LDA. – ttnphns Jan 06 '15 at 08:24
  • http://stats.stackexchange.com/q/158772/3277 similar question – ttnphns Jun 26 '15 at 12:41

0 Answers0