A method of assessing the accuracy and validity of a statistical model. The available data are divided into two parts. Modelling of the data uses one part only. The model selected for this part is then used to predict the values in the other part of the data. A valid model should show good predictive accuracy. See also machine learning.