Predicting how a model's behavior would change if specific training data were excluded without retraining.