0

How to compute False Accept and False reject rates using one class SVM? I have a user data with around 70000 samples. I am trying to apply one class SVM here. The number of -1 values obtained are 12765 and the rest are displayed as 1. From these values obtained, how do I compute False Accept Rate ?

sindhu reddy
  • 61
  • 8

1 Answers1

1

You can compute it with the help of Confusion matrix.

FAR = FPR = FP/(FP + TN)

FRR = FNR = FN/(FN + TP)

where FP: False positive
      FN: False Negative
      TN: True Negative
      TP: True Positive

You can also find the original answer here

Pathi_rao
  • 93
  • 8
  • I understand the formula to compute FAR and FRR and have tried it using binary classifiers. But my question is how do I compute it in one-class classifier ? As mentioned I have number of '-1' values as12000 and the remaining 58000 as '+1' values. – sindhu reddy Jul 28 '21 at 16:15
  • Eg: When you are predicting only "dog class" in your model and if your model *does not* predict correctly, you would still get 2 labels. One for correctly predicting the Dog and the other for predicting it Falsely. Therefore there wouldn't be such thing as a one-class classifier. The single class implies that the answer is always 1 but we shouldn't leave the fact that the answer is not 1. – Pathi_rao Jul 28 '21 at 18:08