A technique in cross-validation where the data is partitioned into k subsets (or "folds"), where the first k-1 folds are used for training and the last fold for evaluation. The process is repeated k times, leaving out a different fold for evaluation each time.