A Multilevel Temporal Context Network for Sleep Stage Classification

Table 6

Comparison between MLTCN and other existing approaches on Sleep-EDF-2013 and Sleep-EDF-2018 datasets.

Dataset	Model	Architecture	Input	Approach	ACC	MF1	Kappa
Sleep-EDF-2013	Phan et al. [28]	CNN	Spectrogram	One-to-one	79.1	69.8	0.70
	Tsinalis et al. [7]	CNN	Raw signal	Many-to-one	74.8	69.8	0.65
	Vilamala et al. [23]	CNN	Spectrogram	Many-to-one	81.3	76.5	0.74
	Seo et al. [22]	CNN + Bi-LSTM	Raw signal	Many-to-one	83.6	76.5	0.77
	Phan et al. [14]	CNN	Spectrogram	One-to-many	81.9	73.8	0.74
	Supratak et al. [25]	CNN + RNN	Raw signal	Many-to-many	82.0	76.9	0.76
	Zhang et al. [16]	DCNN + RNN	Spectrogram + raw signal	Many-to-many	83.8	—	—
	Yang et al. [10]	1D-CNN-HMM	Raw signal	Many-to-many	83.98	76.9	0.78
	MLTCN (ours)	CNN + TCN + HMM	Spectrogram + raw signal	One-to-many	84.2	77.1	0.78
Sleep-EDF-2018	Phan et al. [14]	CNN	Spectrogram	One-to-many	79.6	72.8	0.72
	Supratak et al. [25]	CNN + RNN	Raw signal	Many-to-many	77.8	71.8	0.70
	MLTCN (ours)	CNN + TCN + HMM	Spectrogram + raw signal	One-to-many	81.0	74.9	0.74