Breast Cancer Subtype Specific Classifiers of Response to Neoadjuvant Chemotherapy Do Not Outperform Classifiers Trained on All Subtypes
Figure 1
Cartoon of the double loop cross-validation scheme.
Our analysis employed a double look cross-validation. The inner loop determines the optimal number of features to be used by a specific combination of feature selection and classifier, here depicted by the green block. This inner loop uses 2/3 of all data (i.e. the training data), the remaining 1/3 is employed to measure the performance of the trained classifier (i.e. a 3 fold cross-validation setup). The outer loop is repeated 15 times in order to get an average AUC for each predictor.