0
$\begingroup$

I was doing some tests regarding my PR curve for 2 different models (first image), and I got the idea of ploting the log of those curves (second image) to see if there were any insights that I could take from that.

I failed to find any previous work regarding this, can this new plot be any useful to draw conclusions beyond what the original plot already indicated?

(the red line in the plots indicates the random classifier, the positive class is only 0,22% of the dataset)

Original PR curve

Log PR curve

Improve this question
$\endgroup$
3
  • $\begingroup$ You might do well to spend more time on basic, efficient, measures of accuracy and predictive discrimination (mean squared prediction errors, pseudo $R^2$, etc.). $\endgroup$
    – Frank Harrell
    Commented Dec 16, 2023 at 14:16
  • $\begingroup$ Yes, I already have, I was just wondering if this form of visualization could give any insights since I didn't find anything in the literature. Maybe make the intervals where the blue model is better than the red even clearer? $\endgroup$
    – GabrielPast
    Commented Dec 16, 2023 at 15:29
  • $\begingroup$ I see zero insights coming from that or from ROC curves. Neither are consistent with how predictions are translated into decisions. $\endgroup$
    – Frank Harrell
    Commented Dec 16, 2023 at 17:28

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.