Approaching cross-validation techniques
Cross-validation is a method used in model selection procedures based on the principle of predictive accuracy. A sample is divided into two subsets, of which the first (training set) is used for construction and estimation, while the second (validation set) is used to verify the accuracy of the predictions of the estimated model. Through a synthesis of repeated predictions, a measure of the accuracy of the model is obtained. A cross-validation method is like jackknife, in that it leaves one observation out at a time. In another method, known as K-fold validation, the sample is divided into K subsets and, in turn, each of them is left out as a validation set.
Cross validation can be used to estimate the Mean Squared Error (MSE) test (or, in general, any measure of precision) of a statistical learning technique in order to evaluate its performance or select its level of flexibility.