Verification of predicted robustness and accuracy of multivariate analysis

► Verification of model selection methods for a small sample (42). ► Use of a much larger sample (166) from ADNI for the verification of the PCA/FDA model. ► The .632 bootstrap or CV estimators can fail encouraging unrobust solutions. ► Bootstrap with the proposed metric can well predict robustness of chosen models.► Scanner types and image reconstruction methods had an effect on the verification.