Given an imbalanced data set, how would you train a model to accurately classify images?