Table 3.

Balanced accuracy measured on the entire data set, after being divided into cross-validation folds using the Splitting by Age/Sex and Splitting by Site strategies. We evaluated classification performances when models are trained on combined cortical and subcortical features, cortical thickness, cortical surface area, and subcortical volume. Furthermore, all models were trained/tested without and with ComBat harmonization.

	Cortical + subcortical		Cortical thickness		Cortical surface area		Subcortical volume
	No ComBat	With ComBat	No ComBat	With ComBat	No ComBat	With ComBat	No ComBat	With ComBat
Splitting by Age/Sex
SVM linear	0.616	0.524	0.577	0.504	0.572	0.518	0.602	0.524
SVM rbf	0.639	0.525	0.600	0.515	0.578	0.510	0.619	0.513
SVM + PCA	0.638	0.529	0.601	0.513	0.575	0.518	0.622	0.513
SVM + ttest	0.627	0.515	0.581	0.515	0.567	0.526	0.619	0.521
LASSO	0.612	0.524	0.583	0.499	0.578	0.516	0.596	0.518
Ridge	0.610	0.523	0.585	0.498	0.573	0.515	0.594	0.520
Elastic Net	0.609	0.523	0.584	0.500	0.569	0.517	0.593	0.520
Random Forest	0.613	0.507	0.593	0.514	0.574	0.509	0.611	0.511
Splitting by site
SVM linear	0.512	0.519	0.498	0.495	0.499	0.506	0.506	0.521
SVM rbf	0.513	0.515	0.493	0.499	0.493	0.513	0.503	0.519
SVM + PCA	0.527	0.520	0.502	0.512	0.504	0.524	0.520	0.520
SVM + ttest	0.502	0.512	0.487	0.499	0.507	0.508	0.510	0.527
LASSO	0.513	0.517	0.491	0.489	0.508	0.513	0.507	0.512
Ridge	0.514	0.514	0.497	0.490	0.505	0.509	0.507	0.514
Elastic Net	0.513	0.514	0.498	0.489	0.503	0.514	0.507	0.514
Random Forest	0.518	0.506	0.495	0.501	0.491	0.503	0.519	0.501