In predictive modeling, holdout data is used to evaluate the model’s predictive accuracy.

• A. To train the model
• B. To evaluate performance
• C. To calibrate parameters
• D. To collect more data

Answer: (B)

Holdout data is used to measure how accurately a predictive model will perform on unseen data. By keeping a portion of the data separate from training, you get an unbiased estimate of generalization, then evaluate the model on this holdout set to see how it would perform in the real world. This helps detect overfitting, where a model does well on training data but poorly on new data.

Training the model uses the training portion of the data, not the holdout. Calibrating or tuning parameters is typically done with validation techniques (like cross-validation) on the training data so you don’t bias the holdout evaluation. Collecting more data is about expanding the dataset to improve future performance, not evaluating the current model.