Channel and Feature Selection for a Motor Imagery-Based BCI System Using Multilevel Particle Swarm Optimization

Table 5

Classification results with different levels of PSO for feature selection after channel selection.

Level	SF	ACC (%)	TRT (s)	TTT (s)	TET (s)	Kappa	F-score	SEN	SPE	PRE
Feature selection after one-level PSO for channel selection
1	434	97	2,520	9,778	0.892	0.94	97.03	98	96	96.08
2	215	98	4,024	11,300	0.198	0.96	98	98	98	98
3	125	99	4,629	11,887	0.153	0.98	98.99	98	100	100
4	54	99	4,887	12,145	0.107	0.98	98.99	98	100	100
5	37	99	5,058	12,316	0.103	0.98	98.99	98	100	100
6	28	99	5,299	12,557	0.101	0.98	98.99	98	100	100
7	17	99	5,417	12,675	0.098	0.98	98.99	98	100	100
8	17	99	5,530	12,788	0.098	0.98	98.99	98	100	100
Feature selection after two-level PSO for channel selection
1	263	98	1,443	11,305	0.241	0.96	98	98	98	98
2	122	99	2,106	11,968	0.149	0.98	98.99	98	100	100
3	56	99	2,389	12,251	0.112	0.98	98.99	98	100	100
4	33	99	2,572	12,434	0.102	0.98	98.99	98	100	100
5	25	99	2,703	12,565	0.101	0.98	98.99	98	100	100
6	20	99	2,818	12,680	0.099	0.98	98.99	98	100	100
7	12	99	2,913	12,775	0.095	0.98	98.99	98	100	100
8	12	99	3,005	12,867	0.095	0.98	98.99	98	100	100
Feature selection after three-level PSO for channel selection
1	235	99	1,437	12,546	0.225	0.98	98.99	98	100	100
2	133	99	2,225	13,334	0.163	0.98	98.99	98	100	100
3	66	99	2,586	13,695	0.131	0.98	98.99	98	100	100
4	36	99	2,805	13,914	0.107	0.98	98.99	98	100	100
5	26	99	2,984	14,093	0.103	0.98	98.99	98	100	100
6	16	99	3,136	14,245	0.096	0.98	98.99	98	100	100
7	15	99	3,275	14,384	0.096	0.98	98.99	98	100	100
8	15	99	3,391	14,500	0.096	0.98	98.99	98	100	100
Feature selection after four-level PSO for channel selection
1	196	99	958	14,228	0.189	0.98	98.99	98	100	100
2	107	99	1,451	14,721	0.151	0.98	98.99	98	100	100
3	58	99	1,710	14,980	0.122	0.98	98.99	98	100	100
4	28	99	1,861	15,131	0.101	0.98	98.99	98	100	100
5	18	99	1,962	15,232	0.099	0.98	98.99	98	100	100
6	18	99	2,068	15,338	0.099	0.98	98.99	98	100	100
7	17	99	2,150	15,420	0.097	0.98	98.99	98	100	100
8	17	99	2,236	15,506	0.097	0.98	98.99	98	100	100

SF: selected features; ACC: accuracy; TRT: training time; TTT: total training time of channel selection and feature selection; TET: test time; SEN: sensitivity; SPE: specificity; PRE: precision.