by Michele Laurelli
Data held out for final evaluation of a trained model, never seen during training.
Used to assess real-world performance. Should only be used once at the end. Prevents overfitting to validation set.
20% holdout test set
Final evaluation data
Unseen examples