K-fold cross-validation is a technique used to validate a machine learning algorithm for a particular dataset. The dataset is divided into k pieces (of size N/k, where n is the size of the full dataset) and then the algorithm is trained on k-1 of them with the remaining N/k of the dataset used as test data for an accuracy measure. This is then repeated with a different piece left out and the results of the accuracy measure averaged.
Used on pages 182, 183