I would add some intuition about why this metrics is useful, especially on imbalanced datasets.

If the above comments are addressed, I am +1! Thanks @TTRh !

Thanks for the comments !

• I use confusion matrix to compute balance accuracy as proposed.
• I change weight parameter to balance. I hope it is better, i think it could be great to have this parameter to adapt this score metrics.
• For the user guide part, should i add a paragraph input as accuracy score or confusion matrix or just add words in binary classification section ?

For the user guide part, should i add a paragraph input as accuracy score or confusion matrix or just add words in binary classification section ?

+1 for a separate section.

Any news on this PR? The code seems OK (but I don't have much experience). Two remarks:

• The code should check for special cases like tp == 0 or pos == 0 (maybe using _prf_divide)
• There is a format issue in sklearn/metrics/classification.py on lines 245-246 (missing space around ,)

Hi, sorry i forgot that one, i should also add a separate section in the user guide. I will update this PR regarding comments.

I just added a separate section in user guide. Also i'm thinking about extend balanced accuracy to multiclass classification problems by taking balanced accuracy as the recall for each class averaged over the number of classes. WDYT @arjoly ?
Thanks !

@TTRh Do you have example of such an extension?

Btw, there's been yet another attempt at this enhancement in #6752

Closed by #8066.