1
$\begingroup$

The area under the receiver-operator characteristic curve has a interpretation of how well the predictions of two categories are separated.

This post gives the area under the precision-recall curve as the average precision across all thresholds. This is unsatisfying. Are there other interpretations?

Improve this question
$\endgroup$
3
  • 2
    $\begingroup$ I believe you are basically making the same question as the post you linked $\endgroup$
    – Firebug
    Commented May 10, 2023 at 12:42
  • 1
    $\begingroup$ @Firebug That one seemed to be specific to average precision, while I am looking for other interpretations (which might not exist), though if this gets closed as a duplicate, I won’t go on Meta to request it be reopened. $\endgroup$
    – Dave
    Commented May 10, 2023 at 12:45
  • 2
    $\begingroup$ I think there are not really any other ones. I would also see this in reverse too, as there is no literature associating AUC-RP with any known statistical tests. If there were such tests, it would indicate some additional interpretation. (Of course, the absence of evidence is not the evidence for absence, but I am pretty sure many people have tried to create "that evidence" already.) What I have seen is the Precision-Recall-Gain curve being associated with the expected $F_1$ score but that's "reaching" as it is not really the AUPR curve. $\endgroup$
    – usεr11852
    Commented May 10, 2023 at 13:14

0

Reset to default

Your Answer

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