Table 6.

Performance of classifiers separating HCs from AD patients

Learner	Features	Accuracy	Sensitivity	Specificity	Precision	F-score	AUC
Average performance
RFC	Ling	0.72	0.76	0.67	0.74	0.75	0.72
SVM	Ling	0.75	0.75	0.74	0.77	0.76	0.75
RFC	Cov	0.73	0.78	0.67	0.73	0.75	0.72
SVM	Cov	0.74	0.80	0.67	0.74	0.77	0.74
RFC	Phon	0.59	0.66	0.52	0.62	0.64	0.59
SVM	Phon	0.62	0.70	0.52	0.63	0.66	0.61
RFC	Cov + Ling	0.78	0.84	0.72	0.78	0.81	0.78
SVM	Cov + Ling	0.79	0.79	0.78	0.82	0.80	0.79
RFC	Best∗	0.75	0.78	0.71	0.76	0.77	0.74
SVM	Best∗	0.79	0.81	0.77	0.81	0.81	0.79
Best model
RFC	Ling	0.81	0.77	0.86	0.87	0.82	0.82
SVM	Ling	0.85	0.85	0.86	0.88	0.86	0.85
RFC	Cov	0.85	0.88	0.82	0.85	0.87	0.85
SVM	Cov	0.85	0.88	0.82	0.85	0.87	0.85
RFC	Phon	0.67	0.65	0.68	0.71	0.68	0.67
SVM	Phon	0.72	0.84	0.57	0.70	0.76	0.71
RFC	Cov + Ling	0.94	1.00	0.86	0.90	0.95	0.93
SVM	Cov + Ling	0.88	0.81	0.95	0.95	0.88	0.88
RFC	Best∗	0.85	0.85	0.86	0.88	0.86	0.85
SVM	Best∗	0.87	0.80	0.95	0.95	0.87	0.88

Abbreviations: AD, Alzheimer's disease; AUC, area under the curve of receiver operating characteristics; HCs, healthy elderly controls; RFC, Random Forests Classifier; SVM, Support Vector Machine classifier; Ling, set of all linguistic features; Cov, set of all information coverage features; Phon, set of all phonetic features; Cov + Ling, a combination of all linguistic and information coverage features.

NOTE. The best results are indicated in bold.

^∗A combination of all features with P value < .001 when correlating with cognitive impairment.