hard20 pts · first answer countsclass-imbalancedata-leakage

Oversampling before cross-validation

A colleague applies SMOTE to the whole dataset and then runs 5-fold cross-validation, getting excellent recall. What is wrong?

A.SMOTE only works with neural networksB.Five folds are too few for imbalanced dataC.SMOTE should be applied only to the test foldD.Synthetic points built from validation examples leak into training, so the scores are too optimistic

Discussion 0

Join free to post in the discussion.

Oversampling before cross-validation

Discussion 0

Related questions