Table 2.

Classification performance.

Speech Task (# patients)	Features	Classifiers	Top 5 features
Picture description (25 patients)	NS	RF	0.61 ± 0.05
	SF	RF	0.77 ± 0.04
	NS + SF	EN	0.79 ± 0.07
	MFCC	EN	0.54 ± 0.08
	MFCC + NS	EN	0.50 ± 0.06
	MFCC + SF	LR-l1	0.89 ± 0.06
	MFCC + SF + NS	LR-l1	0.89 ± 0.05
Reverse counting (25 patients)	NS	RF	0.41 ± 0.07
	MFCC	NB	0.84 ± 0.02
	MFCC + NS	RF	0.79 ± 0.03
Diadochokinetic rate (24 patients)	NS	LR-l1	0.53 ± 0.06
	MFCC	NB	0.60 ± 0.07
	MFCC + NS	NB	0.58 ± 0.06

Performance achieved in each task for the different feature sets. Only the classifiers with highest accuracy value are shown. Accuracy is computed as the average (± s.d.) of 50 runs with 10-fold cross-validation. MFCC features are relevant for achieving good performance in the different speech tasks.