Figure 34.

Comparison of model selection techniques for little and big data. For little data, one can use k-fold cross-validation with a separate test set (a) whereas the holdout method with three sets can be used for big data (b). In k-fold cross-validation the data is split into k folds and one loop over all the k-folds, using k – 1 folds as the training set and the kth fold for testing.