A test set is a subset, a part of the available data isolated before building a model, usually between 20 to 30% of the entire dataset. Test sets are used for evaluating the accuracy of models fitted on a training set.