Hello Mainak,

Your ROC-curve should plot true positive rate (Sensitivity) in respect to false positive rate (Specificity), according to the different cut-off points, showing a trade-off between these two characteristics.

For details, please see the following links.

To do this in R:

http://rocr.bioinf.mpi-sb.mpg.de/ROCR.pdf

And in MATLAB:

http://www.mathworks.se/help/toolbox/stats/perfcurve.html

That I understand. While plotting a ROC curve for 1 variable we give the values on one colume and group (denoted by 0 and 1) on the other. But in case of PLSDA model, we are taking care of multiple variables. So, what values should we use?

Ok. Then, there are two options: 1-vs-rest or 1-vs-1.

If you're familiar with R, you can find the second solution implemented in 'pROC' package:

http://cran.r-project.org/web/packages/pROC/pROC.pdf

(Haven't tried this toolbox myself, but it should be easy)

P.S.:

You might have already seen this: http://www.cs.bris.ac.uk/~flach/ICML04tutorial/ROCtutorialPartIII.pdf