Spatial and Time Domain Feature of ERP Speller System Extracted via Convolutional Neural Network

Table 1

Results of the CNN classification. Data are sorted according to the ERP group. Accuracy (Acc.), sensitivity (Sens.), precision (Prec.), F1 measure, ROC, PSNR, and peak time of 2nd layer (PeT.) are given for comparison.

Subject number	Type	Acc.	Sens.	Prec.	F1 measure	ROC	PSNR	PeT.
1	H	.917	.250	.028	.050	.695	−42.285	.372
2	H	1.000	.647	.131	.218	.863	−34.468	.485
3	H	.917	.750	.188	.300	.997	−35.677	.594
4	H	.833	.750	.255	.344	.766	−32.909	.437
5	H	.833	.744	.242	.366	.660	−37.448	.354
6	H	.750	.782	.276	.408	.562	−39.263	.449
7	H	.833	.803	.292	.428	.814	−25.565	.595
8	H	1.000	.826	.317	.458	.873	−25.902	.411
9	H	.667	.844	.333	.478	.696	−22.070	.527
10	H	.750	.838	.346	.490	.873	−39.588	.367
11	H	.917	.869	.327	.475	.922	−25.750	.448
12	H	.917	.878	.342	.493	.940	−24.519	.664
13	H	.667	.747	.294	.422	.638	−23.987	.497
14	H	.833	.713	.279	.401	.778	−40.687	.543
15	H	.917	.721	.290	.414	.935	−39.207	.489
16	H	.750	.733	.302	.428	.998	−35.497	.362
17	H	.917	.733	.289	.415	.799	−38.910	.284
18	H	.917	.740	.295	.421	.861	−27.944	.452
19	H	1.000	.746	.298	.426	.780	−29.202	.458
20	L	.583	.846	.344	.489	.843	−25.722	.445
21	L	.583	.854	.343	.490	.573	−18.743	.575
22	L	.667	.849	.338	.483	.249	−21.219	.341
23	L	.833	.849	.341	.486	.427	−46.836	.638
24	L	.750	.853	.337	.483	.582	−20.236	.282
25	L	1.000	.860	.343	.488	.888	−20.511	.558
26	L	.917	.866	.343	.492	.535	−22.905	.627
27	L	.833	.868	.337	.485	.580	−22.883	.451
28	L	.750	.881	.362	.513	.898	−23.225	.381
29	L	.583	.808	.321	.460	.709	−31.783	.350
30	L	.833	.814	.324	.464	.874	−36.084	.396
31	L	.583	.755	.298	.428	.742	−27.483	.422
32	L	.667	.752	.303	.432	.931	−19.580	.533
33	L	.583	.745	.294	.432	.377	−32.561	.454