Schematic of a three-fold CV process. The original training data are divided into three non-overlapping subsets. Each fold is held out in turn while models are fitted to the remaining training folds. Performance metrics, such as MSE or R², are estimated using the retained fold in a given iteration. The CV estimate of the model performance is given by the average of the three performance estimates.