In the paper one sample is used for
regression and a second for prediction. Mosteller and
Turkey [9], and various other people further developed
the idea. A clear statement of cross-validation, which is
similar to current version of k-fold cross-validation,
first appeared in [8].