TABLE 3.

Comparison of our method to various state-of-the-art motor imagery DA techniques.

Researcher	Augmentation method	Dataset	Classification	Evaluation	Accuracy improvement (%)
Zhang et al. (2018)	Noise introduction in the frequency domain	“2a”	CNN [EEG time series]	20% validation	2.3
Gubert et al. (2020)	Extension of common spectral spatial patterns	“2a”	Fisher-LDA [CSP-based features]	50% validation	5.1
Lee et al. (2021)	Ensemble empirical mode decomposition	“2a”	CNN [Filter Bank CSP]	20% CV	8.2
Fahimi et al. (2021)	Deep convolutional generative adversarial network	“4a” (BCI comp. III)	deep-CNN [EEG time series]	50% validation	3.5
Our work	NFT	“2a”	LDA [CSP TP feature]	67% CV	2.1