Figure 3

The cross-validation procedure. The dataset is divided into three sets: the training set TG, the validation set VS, and the test set TS. Each model hypothesis h _i, i = 1,…, k is first trained on TG; its performance is then evaluated on VS. Based on this performance, the best model (in the figure, h ₂) is then selected, and its performance finally estimated based on its prediction error on the data included in the test set TS.