0
$\begingroup$

The question may look rediculous. The problem in my hand is a multi-class (5 class labels) classification problem and I coded it a year back. I am having the resulsts such as G-mean, F-score, and ROC_AUC score with me, but I lost my code due to the crash of my machine. I can resconstruct the code and get all the necessary details, but I am running short of time to submit the final report. I am in a need of including the ROC curves for different models in my final report. Can I plot the curves having the ROC_AUC scores calculated using the function roc_auc_score() ?

$\endgroup$
3
  • $\begingroup$ I'm not familiar with the G-mean, but I assume it is a single number? $\endgroup$
    – Calimo
    Jul 29 '20 at 6:27
  • $\begingroup$ yes it is a single number $\endgroup$ Jul 29 '20 at 7:02
  • $\begingroup$ Then my answer holds. $\endgroup$
    – Calimo
    Jul 29 '20 at 7:03
3
$\begingroup$

You probably can't.

The ROC AUC score, F-score and G-mean are all summarized measure. This means that information about the specific underlying distributions has been lost: you cannot recover raw data from summarized data.

There are an infinity of ROC curves that have a given AUC. If you know the number of data points, you might be able to restrict the number of possible curves, but you will still have multiple (many) curves to choose from, and no way to know which one was the correct one.

The only exception is if your AUC is exactly 1.0 or 0.0 and you had a perfect classification.

$\endgroup$
1
  • $\begingroup$ Thanks a lot for your answer. My AUC scores are between 0.0 and 1.0. $\endgroup$ Jul 29 '20 at 7:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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