From the book "Data Analysis and Data Mining" by Adelchi Azzalini and Bruno Scarpa, page 206:
Such a low percentage and the small absolute value suggest using cross-validation to trade off bias and variance. We assessed the performance of different models by 10-fold cross-validation, using the same random partition for all methods. To compare the actually observed data, we predicted each of the 10 parts using the best model fitted by using the other parts, thus avoiding having to divide the data set into training and validation sets. We typically also used inner cross-validation to choose a model within each class.
What is inner cross validation?