Performing many learning experiments with the same model architecture and different data sets, we can make proper statistical arguments about the performance of the modeling approach. Cross-validation is a technique for performing many experiments with a small data set.