When working with imbalanced data, judging your model's performance using accuracy is like pouring diesel on a burning flame, if not careful you too might burn. Avoid accuracy! Please use F1-score, However, depending on your problem definition you call settle for high precision or recall.
Note: You can't have high recall and precision at same time. You can only have either of both.
I’d like to know what your answer is as well? 😉