Should using training datasets or testing datasets for evaluating the performance of the models

Question

When comparing multiple models, should use training datasets or testing datasets or both of training and testing datasets for evaluating the performance of the models?

Answer 1

You can not rely on the training dataset because it was used to build the model. You have to test the model on an unseen dataset. For example, it is not recommended to select the model which perform well on the training dataset but it performs badly on the testing (overfitted).