2
$\begingroup$

I have compared the rankings obtained by comparing 10+ classifiers with this two metrics:

  • Jaccard score
  • F1 score

They show a perfect correlation. This results holds on 50+ datasets. When comparing with other metrics, it's clear that there are variations in the rankings produced.

Is there something in the respective definitions of this two metrics that ensure that they produce the same rankings of classifiers? In other words, is there a monotone relation between the two?

$\endgroup$

1 Answer 1

3
$\begingroup$

I confirm that F1 score and Jaccard score are monotonically related in the sense that they produce the same orderings of models.

$$ J = \frac{TP}{FN + FP + TP} $$

$$ F_1 = \frac{TP}{\frac{1}{2}(FN + FP) + TP} $$

More details can be found here: F1/Dice-Score vs IoU

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.