Practical Predictive Analytics

Training/test/validation datasets

Training data is the data that is used to build the predictive data. Test data, also known as a hold-out sample, is the data which was not used in the training process. Test data is used to see if a trained model is able to generalize its results to another dataset which was not part of the model training.

However, even though the test dataset is not used in the training of the model, the contents are usually viewable by the modeler during model development.

A validation set is a third dataset which is sometimes used to further benchmark the model results. Validation data is also data that has never been seen before, which is incorporated after the model has passed validation against the test dataset and ...