measures that evaluate how well an estimated model will perform in an unseen sample, rather than making the evaluation solely on the basis of the sample data used to build the model.