A Novel Approach for Sleep Arousal Disorder Detection Based on the Interaction of Physiological Signals and Metaheuristic Learning

Table 5

The Lyapunov exponent descriptor was used to calculate accuracy, AUC, and RR in various scenarios of the subset feature selection. The results of SVM and improved SVM classifiers are compared to detect arousal events.

Signal type	No. of features	SVM			Optimized SVM
		ACC (%)	AUC (%)	RR (%)	ACC (%)	AUC (%)	RR (%)
EEG	8	73.03	73.60	71.54	80.47	80.99	80.76
	12	78.51	79.10	79.83	83.35	84.71	83.02
	16	74.98	75.75	75.92	84.27	83.54	81.43
	20	76.15	74.33	76.78	82.72	82.07	81.11
ECG	8	71.34	71.86	71.04	80.19	80.14	80.24
	12	76.23	76.07	76.66	82.74	83.05	82.62
	16	73.45	73.65	73.69	81.73	81.57	80.68
	20	72.14	72.62	72.87	80.78	79.66	79.11
EEG + ECG	8	76.54	76.46	76.75	90.14	89.23	90.80
	12	79.06	78.81	79.71	91.42	91.70	91.56
	16	77.80	77.67	77.12	91.87	91.46	90.48
	20	73.11	73.24	73.49	90.93	90.83	89.31

Bold values are the best values obtained.