Train / test split

To know a model learned rather than memorised, test it on data it never saw in training.

Split the data, commonly 80% train / 20% test. Learn on the training set; measure on the held-out test set. Great-on-train but poor-on-test means overfitting.