I've recently see more people using "balanced accuracy" for imbalanced binary and multi-class problems. I think it is the same as macro average recall. If so, I think we might want to create an alias, because it is not super obvious, and maybe add a scorer.
Also see EpistasisLab/tpot#108