Nested hyperparameter tuning for the regularization parameter in linear SVM.

For each disorder, we compared the cross-validated classification performance of each model (e.g., left entorhinal cortex in Aregion) using either the fixed regularization parameter (C = 1) or by performing nested cross-validation within each held-out training set to optimize the C parameter based on balanced accuracy. The mean balanced accuracy is indicated for each model type with versus without tuning the C parameter. (CSV)