I was checking if it is possible to get per class accuracy from confusion matrix until I found a similar question in SO.
Unfortunately, the answer that to me seems convincing gets negative votes:
Accuracy is a global measure, and there is no such thing as class-wise accuracy.
and the highly upvoted one is followed by hot arguments such as
This calculates not accuracy, but recall
in the comments section. No accepted answer.
I would like to have a YES or NO answer to this question. Is it technically correct (and possible) to get model accuracy per class from confusion matrix?