A Novel Time-Incremental End-to-End Shared Neural Network with Attention-Based Feature Fusion for Multiclass Motor Imagery Recognition

Table 2

Comparison of the accuracy and kappa coefficient between the method in this paper and the methods in the literature.

Methods	Statistics	Subjects									Mean ± sd
		A01	A02	A03	A04	A05	A06	A07	A08	A09
Tabar and Halici [31]	Accuracy kappa	64.0%; 0.520	49.7%; 0.330	49.7%; 0.330	92.5%; 0.910	74.5%; 0.660	68.5%; 0.580	61.7%; 0.490	62.5%; 0.500	59.5%; 0.460	66.2 ± 11.2%; 0.550 ± 0.152
Ang et al. [47]		76.0%; 0.680	56.5%; 0.420	81.3%; 0.750	61.0%; 0.480	55.0%; 0.400	45.3%; 0.270	82.7%; 0.770	81.3%; 0.750	70.7%; 0.610	67.8 ± 12.9%; 0.570 ± 0.173
Winkler et al. [48]		69.3%; 0.590	55.7%; 0.410	86.5%; 0.820	67.7%; 0.570	53.5%; 0.380	46.7%; 0.290	84.2%; 0.790	85.0%; 0.800	79.0%; 0.700	69.7 ± 14.2%; 0.600 ± 0.189
Nicolas-Alonso et al. [49]		87.2%; 0.830	63.3%; 0.510	91.0%; 0.880	76.0%; 0.680	67.0%; 0.560	51.2%; 0.350	92.5%; 0.900	88.0%; 0.840	89.5%; 0.750	78.4 ± 14.0%; 0.700 ± 0.180
Ai et al. [50]		82.7%; 0.770	65.5%; 0.540	88.0%; 0.840	77.5%; 0.700	72.2%; 0.630	70.7%; 0.610	82.7%; 0.770	88.0%; 0.840	89.5%; 0.860	80.2 ± 8.10%; 0.730 ± 0.108
Nicolas-Alonso et al. [51]		88.0%; 0.840	66.2%; 0.550	92.5%; 0.900	78.2%; 0.710	74.5%; 0.660	58.0%; 0.440	95.5%; 0.940	88.7%; 0.850	82.0%; 0.760	80.4 ± 11.8%; 0.740 ± 0.157
Xu et al. [52]		76.7%; 0.690	68.5%; 0.580	100%; 1.000	100%; 1.000	87.2%; 0.830	85.0%; 0.800	78.2%; 0.710	71.5%; 0.620	73.0%; 0.640	81.2 ± 28.5%; 0.750 ± 0.147
Zhang et al. [16]		87.7%; 0.850	65.5%; 0.540	88.2%; 0.870	83.5%; 0.780	82.7%; 0.770	74.5%; 0.660	93.2%; 0.920	87.2%; 0.830	91.5%; 0.900	83.0 ± 8.34%; 0.800 ± 0.120
SCNN-BiLSTM network based on attention		88.2%; 0.840	73.3%; 0.650	79.7%; 0.730	89.2%; 0.870	80.6%; 0.740	76.4%; 0.690	90.3%; 0.870	81.6%; 0.760	85.4%; 0.810	82.7 ± 5.57%; 0.780 ± 0.074