Ensemble Deep Learning for Biomedical Time Series Classification

Table 1

Contribution of subview prediction (explicit training).

Dataset	Method	Sp (%)	Se (%)	GMean (%)	Acc (%)	AUC	NPV = 95%
							TPR (%)	FPR (%)
DS1	Simple [6]	88.84	76.95	82.68	85.41	0.9034	63.322	8.217
	Subview	89.85	76.81	83.07	86.09	0.9123	71.111	9.227
DS2	Simple [6]	88.63	79.55	83.97	85.99	0.9174	74.389	9.558
	Subview	89.92	80.05	84.84	87.05	0.9272	78.879	10.135
DS3	Simple [6]	86.58	77.69	82.01	84.03	0.8972	65.597	8.599
	Subview	87.81	78.03	82.77	85.01	0.9074	72.506	9.505
DS4	Simple [6]	82.75	84.81	83.77	83.91	0.9096	0.091	0.011
	Subview	83.67	85.09	84.38	84.47	0.9153	0.051	0.004
DS5	Simple [6]	79.52	86.20	82.79	83.23	0.9084	0	0
	Subview	80.70	86.64	83.62	84.00	0.9144	0	0
DS6	Simple [6]	81.98	84.90	83.43	83.57	0.9101	0.025	0.003
	Subview	82.80	85.49	84.13	84.26	0.9169	0	0
DS7	Simple [6]	77.81	84.71	81.19	81.28	0.8905	0.010	0.001
	Subview	78.60	85.17	81.82	81.90	0.8964	0	0
DS8	Simple [6]	78.31	84.74	81.47	81.71	0.8913	0	0
	Subview	79.46	85.23	82.30	82.51	0.8976	0.020	0.003
DS9	Simple [6]	83.97	75.40	79.57	81.48	0.8661	1.196	0.159
	Subview	84.62	76.04	80.22	82.13	0.8778	52.160	6.722

We can change the discrimination threshold from 0 to 1 and calculate the corresponding values of Se, Sp, and NPV. As for “0,” it means that the condition of NPV being equal to 95% cannot be satisfied.