Cross-validation is the standard procedure for estimating how well a model will perform on data it has not seen, using only the data you already have. The “Machine Learning Basics” chapter of Goodfellow, Bengio, and Courville’s “Deep Learning” presents it as a basic tool for model evaluation when data is limited.
The most common form is k-fold cross-validation. The dataset is split into k equal parts, or folds. The model is trained on k-1 of them and tested on the one left out, and this is repeated k times so that each fold serves once as the test set. Averaging the results gives a more stable and trustworthy estimate of performance than a single train/test split, especially when data is scarce and a single split might be lucky or unlucky. The extreme case, leave-one-out cross-validation, uses a single example as the test set each time.
Cross-validation is used both to estimate a model’s true accuracy and to choose between options - which algorithm, how much regularization, what hyperparameter settings - by picking whatever generalizes best across the folds. Its cardinal rule is that the test data must never influence training, or the estimate becomes optimistic and misleading.
Why business readers should care: cross-validation is the discipline that separates a model’s real, repeatable performance from a number that just looked good on one lucky split of the data. It is the basic honesty check before any model is trusted in production.