In the questions previous to question 5 when it was asked if one model performed better than another model there was a clear target on a specific metric that was given. For question 5 looking at the given reasoning for the answer a different target and method was used with no clear sign that you are supposed to use that method.
True, we did not mention to still use the balanced accuracy. We could make it explicit. Thanks for reporting.
Improve the question in FIX make question more explicit in ensemble · INRIA/scikit-learn-mooc@c9c8878 · GitHub
This would require a change in FUN
It has been also fixed in FUN.
Along the same lines, the definition of “better” in this question no longer requires an improvement ≥ 0.04, but simply A > B. Using the definition used elsewhere in the quiz the answer should be equivalent performance.