Fig. 4: Overview of the 10-fold nested cross-validation procedure.
From: FASDetect as a machine learning-based screening app for FASD in youth with ADHD
The data are randomly split into 10 stratified folds where one fold is held out as a test set (blue colour). For each split, the 9 folds are split again into 10 folds, with one fold (green colour) to validate the hyperparameters. The hyperparameters with the best average ROC AUC on the validation sets are used to fit the machine learning pipeline on the complete training set (i.e., the 9 outer folds framed in red colour) and tested against the test set (blue colour), resulting in 10 ROC AUC scores.
