Improving model performance with meta-learning