Cross validation provides a more reliable estimate of a machine learning model’s ability to perform on data it does not have access to in the training process. Evaluating a model’s performance on the same data in which it is trained has inherent bias and is not guaranteed to be representative of its ability to generalize to new data.