The classification results from the one-shot model approach.

For each disorder, within each train/test split (out of 10 repeats of 10 train/test splits), the top-performing model was identified per representation type within the training fold. The top-performing model (identified within the training data) was then evaluated on the held-out test fold for the corresponding split, with this out-of-sample balanced accuracy reported. In the case of AFC and AFC_combo, the top-performing model was identified with an intermediate PCA within the training fold to reduce the dimensionality, though the out-of-sample performance was evaluated in the full feature space.
(CSV)