Hold-out – is where one splits up a dataset into a ‘train and ‘a test’ set. The test set is the one used to see the best functionality of the model on unseen data, while the training set is what the model is trained on. A typical split when using the hold-out technique is using 80% of the data training and 20% for testing.
Cross-validation – also known as the ‘K-fold cross-validation, is whereby the set of data is randomly split up into ‘k’ groups. Amongst the groups, one is used for the test set, while the others are used as the training set. The training set trains the model while the test set scores it, and the process should be repeated until each unique group has been used as the test set. For example, for 10-fold cross-validation, the data set will be split into ten groups, and the model will be trained and tested ten separate times for every group to have a chance to be the test set.
Comparison of Hold-out and Cross-Validation
Cross-validation is mostly the preferred technique. It provides the model with the opportunity to train multiple train-test splits, giving a better indication of how to fit a model that will perform on unseen data. On the other hand, the hold-out method depends on one train-test split, making it score depending on how the data is divided into train and test sets.