Table 5.

Imagined speech classification methods summary (continuation).

References	Task	Methods	Brain area and waves	Performance
Cooney et al. (2019)	Classification of /a/, /e/, /i/, /o/, and /u/	- CNN	- All areas	−34% Average accuracy
		- Transfer learning	−2–40 Hz bandpass
Sharon and Murthy (2020)	Binary classification for presence of C/V, ± Nasal, ± Bilab,± /uw/, ± /iy/	- CorrNet	- All areas	C/V: 89% Accuracy
			−1-50 Hz bandpass	± Nasal: 76% Accuracy
				± Bilab: 75% Accuracy
				± /uw/: 82% Accuracy
				± /iy/: 80% Accuracy
Bakhshali et al. (2020)	Binary classification for presence of C/V, ± Nasal, ± Bilab,± /uw/, ± /iy/	- Riemannian distance	- Broca's area and Wernicke's area	C/V: 86% Accuracy
	Binary Classification of /pat/, /pot/, /gnaw/, and /knew/	- Correntropy Spectral Density	−1–50 Hz bandpass	± Nasal: 72% Accuracy
				± Bilab: 69% Accuracy
				± /uw/: 84% Accuracy
				± /iy/: 75% Accuracy
				± Word binary classification: 69% Average accuracy
Pawar and Dhage (2020)	Muti-class and binary classification of “left,” “right,” “up,” and “down”	- Kernel ELM	- All areas	Multi-class: 49% best accuracy
		- Statistical features	- Prefrontal cortex, Wernicke's area, right inferior frontal gyrus, Broca's area	Binary: 85% best accuracy
		- DWT	- Prefrontal cortex, Wernicke's area, right inferior frontal gyrus, Broca's area, primary motor cortex
		- ICA	−0.5–128 Hz bandpass