I’m looking for the best practice to avoid data leakage. I have 1 feature that requires mode imputation. The model is XGBoost Classifier.