The validation set is used both for early stopping and for reporting/comparing results.
This causes validation leakage, since the model is effectively tuned on the validation data.
As a result, the reported validation performance is optimistically biased and likely higher than true generalization performance.
Suggestion:
Use a separate test set for final evaluation, or apply nested cross-validation.
The validation set is used both for early stopping and for reporting/comparing results.
This causes validation leakage, since the model is effectively tuned on the validation data.
As a result, the reported validation performance is optimistically biased and likely higher than true generalization performance.
Suggestion:
Use a separate test set for final evaluation, or apply nested cross-validation.