A percentage of training data that is reserved for use in measuring the accuracy of the structure of the data mining model.