0

I want to plot ROC for Leave-one-out (LOO) cross-validation (CV). I have the decision values as well as probability values for each class from SVM classifier. I did see the answer to this post; however, the answer was not complete about plotting ROC according to probabilities.

As an example the output of my classifier for one sample is like this:

c1a
  1 
attr(,"decision.values")
          1/0
c1a 0.1338180
attr(,"probabilities")
            1         0
c1a 0.6824716 0.3175284
Levels: 0 1
Community
  • 1
user4704857
  • 502
  • 3
  • 12
  • @user777 I have 120 samples not one sample. In my post, I just have shown the output of SVM for one sample. – user4704857 Feb 23 '16 at 14:24
  • @user777 How? That is exactly my problem. – user4704857 Feb 23 '16 at 14:46
  • No, my question is how I can calculate FPR and TPR from these probabilities to plot ROC? – user4704857 Feb 23 '16 at 14:51
  • 1
    Answered here. The only difference is that you specify you're using LOOCV but that's irrelevant to the procedure. http://stats.stackexchange.com/questions/145566/how-to-calculate-area-under-the-curve-auc-or-the-c-statistic-by-hand – Sycorax Feb 23 '16 at 14:54

0 Answers0