Table 3.

Area under the ROC curve for each classifier architecture and feature matrix

		Animal fluency
	NB	RF	SVM	Ensemble avg.	Ensemble log.
Type - Token (train)	0.9674	0.9759	0.9554	0.9708	0.9758
Type - Token (test)	0.9612*	0.9500	0.9512	0.9623*	0.9559
TF-IDF (train)	0.6128	0.9769	0.9558	0.9495	0.9744
TF-IDF (test)	0.5938	0.9636	0.9525*	0.9421	0.9611
PO (train)	0.7229	0.9807	0.9528	0.9638	0.9800
PO (test)	0.7196	0.9646	0.9415	0.9475	0.9638*
TF-IDF + PO (train)	0.7314	0.9809	0.9568	0.9651	0.9803
TF-IDF + PO (test)	0.7219	0.9649**	0.9430	0.9495	0.9638*
		Letter F fluency
Type - Token (train)	0.9606	0.9694	0.9417	0.9634	0.9688
Type - Token (test)	0.9544*	0.9425	0.9424*	0.9550	0.9393
TF-IDF (train)	0.5008	0.9687	0.9409	0.9585	0.9686
TF-IDF (test)	0.5012	0.9547	0.9404	0.9504	0.9531
PO (train)	0.8764	0.9729	0.9479	0.9672	0.9726
PO (test)	0.8821	0.9576	0.9357	0.9552*	0.9559
TF-IDF + PO (train)	0.6649	0.9732	0.9500	0.9632	0.9730
TF-IDF + PO (test)	0.6652	0.9579**	0.9398	0.9516	0.9568*

NB=naïve Bayes, PO=phonological overlap, RF=random forest, SVM=support vector machine, TF-IDF=term frequency-inverse document frequency.

^*Indicates the highest test set AUC for each classifier/ensemble in each verbal fluency task.

^**Indicates the highest test set AUC among all classifiers/ensembles in each verbal fluency task.