Jeremy said in the lecture that to build a feature importance with random forest, our model does not have to be perfect, it is enough for it to be slightly better than a mean prediction. But what is the threshold there ? Is R^2 of 0.2 is good enough? Or Is 0.6 accuracy enough?
I want to learn about it, I searched in the forum but couln’t find it, if anyone can help I would be very appreciated
(Actually I think the question can also be applied calculating feature importance with other models too, what is the minimum acceptiple R^2 or accuracy of our model for us to trust feature importance)