Table 2.

Performance comparison of classification of six motor events using the GAL dataset [34] based on the area under ROC (AUROC).

Method	Pre-Processing	AUROC
Proposed (CSP–CWT)	Preprocessing: CSP filtering and CWT CNN architecture: GoogLeNet 22-layer 2D CNN Evaluation: AUROC for every event from 12 subjects Number of Channel: six-channel (C3, Cz, C4, P3, Pz and P4)	0.985
Singhal, et al. (2019) [5]	Preprocessing: Butterworth Low Pass Filter CNN architecture: five-layer 1D CNN Evaluation: Average AUROC of 12 subjects Number of Channel: 32-channel	0.910
Várszegi (2016) [28]	Preprocessing: Artifact rejection and normalization CNN architecture: six-layer 1D CNN Evaluation: Average AUROC of 12 subjects Number of Channel: 32-channel	0.829