0

I am working on performance measuring and trying to draw a ROC curve, however to draw ROC curve i need TPR and FPR.

As we know,

False Positive Rate (FPR) = FP / (FP + TN)

I have got values of TN and FP both equal to 0, so how can i calculate FPR for this case and put in ROC curve?

Md. Nahiduzzaman Rose
  • 442
  • 2
  • 9
  • 23

1 Answers1

6

First of all

False Positive Rate (FPR) = FP / (FP + TN)

thus

I have got values of TP and FP both equal to 0

is not a problem, as TP is not used in this equation. The only problem would be for FP + TN to be 0, but this is impossible since FP + TN = Negatives (all samples with negative label, no matter how you classify them). Consequently the only case then FPR is undefined is if your dataset does not have negative samples, and then there is no point in binary classification either way.

lejlot
  • 64,777
  • 8
  • 131
  • 164
  • Sorry, I have edited my question TP supposed to be TN, thank you – Md. Nahiduzzaman Rose Sep 16 '16 at 19:29
  • this means that your dataset does not contain **any** negative samples, thus it is not a valid binary labeled dataset. You either compute TN / FP incorrectly, or your data is invalid. – lejlot Sep 16 '16 at 23:06
  • Can please clear my confusion about detecting True Negative in this frame? @lejlot http://stackoverflow.com/questions/39645367/how-to-classify-true-negative-from-a-video – Md. Nahiduzzaman Rose Sep 22 '16 at 17:35