Table 2.

Performances of reported sleep staging using the signals from wearable devices

Signals / Models		# of Stages	# of Rec.	Performance		Ref.
				Agree. (%)	Kappa
EEG (O1, O2, Fpz, F7, F8), Acc. PPG, SpO2 / LSTM		5	25	83.5	0.75	[21]
EEG (F4, C4, O2) / MLP neural network		5	154	89	-	[22]
EEG (Fpz-Cz, Pz-Oz) / CNN		5	20	86	-	[23]
EEG (Fpz-Cz) / CNN + BiRNN		5	61	84.3	0.79	[24]
EEG (F4-EOG) / CNN + LSTM		5	62	86.2	0.80	[25]
EEG (Fpz-Cz) / LSTM		5	12	86.7	-	[26]
EEG (In-ear)	Support Vector Machine	5	16	74.1	0.61	[27]
	Random Forest	5	80	80.6	0.73	[28]
Acc.	Generalized Estimating Equation	2	77	86.3	-	[29]
	Rule-based model	2	228	84.0	-	[30]
ECG	LSTM	4	584	71.9	0.57	[31]
	LSTM	4	584	77	0.61	[32]
	LSTM	5	373	71.2	0.52	[33]
	CNN	4	993	77	0.66	[34]
PPG	Bayesian linear discriminant	4	215	59.3	0.42	[35]
	LSTM	4	60	69.8	0.55	[31]
	CNN + Gated Recurrent Unit	5	894	64.1	0.51	[36]
PPG, Acc. / MLP		3	219	72.3	0.28	[37]
PAT, PPG, Acc. / Rule-based model		4	227	66.0	0.48	[38]

# of Rec. number of data points used in the study, Agree. agreement, Ref. reference, EEG electroencephalogram, Acc. Accelerometry signals, ECG electrocardiogram, PPG photoplethysmogram, Skin Temp. Skin temperature, PAT Peripheral arterial tone, LSTM Long-Short Term Memory, MLP multilayer perceptron, CNN convolutional neural network, BiRNN Bidirectional recurrent neural network, SVM support vector machine