0
$\begingroup$

For a multi-class model, there are always chances that the model is learning one class's features more than the other. But how do I find which class has been weakly learned? Please help.

Improve this question
$\endgroup$
5
  • $\begingroup$ Find the class that is poorly predicted by the model, for example, by finding the class with the largest loss values or highest error rates. $\endgroup$
    – Sycorax ♦
    Commented Nov 14, 2022 at 3:30
  • $\begingroup$ thanks, but how do I find the class with largest loss values? do having TP, FP, TN, FN values for each class play any role to the calculation? Or are there other straightforward methods? @Sycorax $\endgroup$
    – shey
    Commented Nov 14, 2022 at 3:46
  • 1
    $\begingroup$ Have a look at the confusion matrix and look for off-diagonal entries. $\endgroup$
    – cdalitz
    Commented Nov 14, 2022 at 12:49
  • $\begingroup$ Compute the loss individually for each observation, then sum by class. $\endgroup$
    – Sycorax ♦
    Commented Nov 14, 2022 at 14:27
  • $\begingroup$ @cdalitz your answer did the job for me, thank you. $\endgroup$
    – shey
    Commented Dec 6, 2022 at 18:36

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.