What is the primary goal of cross-validation?

• A. To maximize training accuracy on the given data.
• B. To estimate model performance on unseen data by averaging performance across folds.
• C. To minimize data usage.
• D. To select features.

Answer: (B)

Cross-validation aims to provide an honest estimate of how a model will perform on new, unseen data. It does this by splitting the data into several folds, training on the other folds and evaluating on the held-out fold, and repeating this process across all folds. The final performance is the average of those evaluations, which smooths out variability from any single train-test split and gives a more realistic measure of generalization.

This approach helps guard against overfitting to the training data, because the evaluation genuinely tests the model on data it wasn’t trained on. It also makes efficient use of the available data by cycling through different train/test partitions.

The other ideas aren’t the main aim: maximizing training accuracy on the given data can mislead you into thinking the model is better than it will be on new data, since it’s evaluated on data it already saw. Minimizing data usage isn’t the goal, since cross-validation intentionally uses multiple splits to make a robust estimate. Feature selection can be informed by cross-validation, but the primary purpose of cross-validation is to estimate generalization performance, not to select features.