Sure I do, Thank you Sir.

It depends a little on the approach. Sometimes it is helpful to run a set of multivariate analyses to get a first idea.

• Having a dataset and no idea, one could use a Principal Component Analysis (PCA) and check the Eigenvektors of the variable covariance matrix. It is that what you plot as the arrows in the combined PCA/Biplot. The longer the arrows on the first axes, the more important the variable is in discrimination. This, of course, is just an option for datasets where more or less direct relationships among variables and clusters can be found.
• Another option could be the use of a random Forest model and have a look at the Gini Index of the variables. The advantage is, that even non linear relationships can (but not necessarily) be detected.
• Similar things can be done with several other approaches. These often have the option to identify the relevance of individual variables in discrimination.

If interested: I once wrote a short tutorial on how to use some multivariate approaches in R.

Jan

Sorry, forgot to upload.

Attached you find a doc for some students.

It is not finished and not free from flaws, but may give you an idea...

Jan

No problem Jan. thanks for sharing