What are the tips and tricks to improve our auc_roc_score? Example: is balanced data required? is recall is more important than precision? is oversampling is usually better
