True, training and test data are usually variate but still we can't say that it wasn't possible. Perhaps in this particular case training data and variavles used were not sufficient