Fig. 2.

Schematic representation of nested CV. Nested CV involves a secondary inner CV loop using the training data from the primary outer CV split, where different sets of hyperparameters are tested (eg, different values for the C parameter for SVM). The best-performing hyperparameters among the 10 inner folds are then used to train a model in the whole training set defined by the outer loop. This model is then tested using the test set of the outer loop. The final performance is estimated by averaging accuracies in the test set across all 10 outer folds. CV, cross-validation; SVM, support vector machine.