Table 4.

Classification performance per measure per classifier (CNN from scratch and tuned Vgg16 with data augmentation) after 100 training epochs.

Feature	Scratch CNN						Tuned Vgg16 with data augmentation
	ACC	Prec	Recall	F1	kappa	AUC	ACC	Prec	Recall	F1	kappa	AUC
SPEC	0.68	0.67	0.71	0.69	0.37	0.68	0.76	0.72	0.85	0.78	0.52	0.76
Chroma	0.55	0.56	0.53	0.54	0.11	0.55	0.63	0.65	0.56	0.61	0.27	0.63
MFCC	0.71	0.75	0.64	0.69	0.42	0.71	0.61	0.63	0.54	0.58	0.23	0.61
MelSpectrum	0.74	0.70	0.84	0.76	0.48	0.74	0.69	0.68	0.72	0.70	0.38	0.69
PowerSPEC	0.70	0.68	0.75	0.71	0.40	0.70	0.69	0.76	0.54	0.64	0.38	0.68
RAW	0.56	0.57	0.49	0.53	0.13	0.56	0.58	0.56	0.69	0.62	0.16	0.58
Tonal	0.53	0.54	0.59	0.56	0.08	0.54	0.49	0.49	0.70	0.58	− 0.02	0.49
ALL features	0.62	0.63	0.55	0.59	0.23	0.62	0.63	0.62	0.67	0.64	0.26	0.63