What is overfitting and how does cross-validation help mitigate it?

• A. Cross-validation is used to evaluate model on the same data used for training.
• B. Cross-validation increases training data usage by 50%.
• C. Overfitting is when a model fits noise; cross-validation estimates generalization performance and helps select model.
• D. Overfitting is underfitting; cross-validation increases training accuracy.

Answer: (C)

Overfitting happens when a model learns noise in the training data and ends up with high training accuracy but poor performance on new, unseen data. Cross-validation helps mitigate this by giving a more reliable estimate of how the model will generalize. It splits the data into multiple train/validation folds, trains on the majority of the data, and evaluates on the held-out portion in each fold. By averaging performance across all folds, you get a robust measure of generalization that isn’t tied to a single split.

This approach also supports model selection and hyperparameter tuning: you compare configurations by their cross-validated performance and choose the one that generalizes best across folds, rather than optimizing for a single split that might reflect noise.

Why the other ideas don’t fit: evaluating on the same data used for training would mask overfitting, since there’s no true unseen data. Cross-validation doesn’t simply increase training data by a fixed amount; in each fold you train on most of the data and validate on a separate portion, so the effective validation is on unseen data. Finally, overfitting is about learning noise, not underfitting, and cross-validation targets generalization rather than just increasing training accuracy.