Table 4.

Diagnostic performances of strategies trained with 80 % of the total dataset (epoch size = 150).

	Strategy	Fold 1 (n = 925)	Fold 2 (n = 925)	Fold 3 (n = 925)	Fold 4 (n = 925)	Fold 5 (n = 926)	Overall (n = 4626)
AUC	1. Original	0.7248	0.8123	0.7982	0.8595	0.7814	0.7952
	2. Class weight (1:0)	0.7236	0.8071	0.8045	0.8478	0.7897	0.7945
	3. × 10 data (positive only)	0.7687	0.8228	0.8458	0.8568	0.7838	0.8156
	4. × 10 data (all)	0.7479	0.8135	0.8446	0.8677	0.7528	0.8053
Sensitivity	1. Original	0.1750	0.4000	0.2500	0.3250	0.2875	0.2875
	2. Class weight (1:0)	0.4250	0.3750	0.6000	0.3125	0.2375	0.3900
	3. × 10 data (positive only)	0.5125	0.6000	0.6750	0.6500	0.3625	0.5600
	4. × 10 data (all)	0.1625	0.3125	0.2625	0.2500	0.2125	0.2400
Specificity	1. Original	0.9882	0.9751	0.9645	0.9870	0.9775	0.9785
	2. Class weight (1:0)	0.8533	0.9503	0.8568	0.9834	0.9846	0.9257
	3. × 10 data (positive only)	0.8686	0.8911	0.8249	0.8911	0.9610	0.8873
	4. × 10 data (all)	0.9834	0.9846	0.9751	0.9976	0.9894	0.9860
PPV	1. Original	0.5833	0.6038	0.4000	0.7027	0.5476	0.5675
	2. Class weight (1:0)	0.2152	0.4167	0.2840	0.6410	0.5938	0.4301
	3. × 10 data (positive only)	0.2697	0.3429	0.2673	0.3611	0.4677	0.3418
	4. × 10 data (all)	0.4815	0.6579	0.5000	0.9091	0.6538	0.6405
NPV	1. Original	0.9267	0.9450	0.9314	0.9392	0.9355	0.9356
	2. Class weight (1:0)	0.9400	0.9414	0.9577	0.9379	0.9318	0.9418
	3. × 10 data (positive only)	0.9495	0.9592	0.9640	0.9641	0.9410	0.9556
	4. × 10 data (all)	0.9254	0.9380	0.9332	0.9336	0.9300	0.9320
Accuracy	1. Original	0.9178	0.9254	0.9027	0.9297	0.9179	0.9187
	2. Class weight (1:0)	0.8162	0.9005	0.8346	0.9254	0.9201	0.8794
	3. × 10 data (positive only)	0.8378	0.8659	0.8119	0.8703	0.9093	0.8590
	4. × 10 data (all)	0.9124	0.9265	0.9135	0.9330	0.9222	0.9215

AUC, area under the curve; PPV, positive predictive value; NPV, negative predictive value.