Model card metric labels: Accuracy or balanced accuracy?

#2
by Alexcchristou - opened

Could the evaluation metadata use “Balanced accuracy” for the entries taken from Table 2 of the paper?

For example, the README metadata lists TUAB = 0.8315 and BCICIV2a = 0.6396 with both metric type and name set to “Accuracy”. These match the REVE-Base values in Table 2, whose caption specifies balanced accuracy.

Current metadata · Paper, Table 2

Making the metric explicit would help readers compare results on datasets with uneven class counts. I have checked the card and table, not rerun the evaluations; this is a labelling question, not a challenge to the reported scores.

Broader Artificial Intelligence org

Hello,
It was changed, thanks for the suggestion.

jonathan-lys changed discussion status to closed

Sign up or log in to comment