Table 4.

Accuracy comparison of the proposed method with other multi-level drowsiness classification methods based on physiological signals.

Study	Method	Physiological Signal	Variable	Accuracy
Barua et al., 2018 [58]	Support vector machine	EEG and EOG	EEG-PSD, blink duration, and contextual analysis	79% (3-level)
Lemkaddem et al., 2018 [60]	Support vector machine	PPG	HRV and PERCLOS	89% (3-level)
Persson et al., 2020 [22]	Random forest classifier	ECG	HRV	64% (3-level)
Arefnezhad et al., 2022 [59]	CNN	ECG	Scalogram of ECG	79% (3-level)
Hultman et al., 2021 [53]	CNN-LSTM	ECG and EOG	Pre-processed ECG and EOG signals	46% (5-level)
Jeon et al., 2019 [54]	Deep spatiotemporal convolutional bidirectional LSTM network (DSTCLN)	EEG	Pre-processed EEG signal	69% (5-level)
This paper	CNN-LSTM	ECG and respiration	HRV, HRV-PSD, and respiration rate	91% (3-level) 67% (5-level)

EEG: electroencephalogram, EOG: electrooculogram, PPG: photoplethysmography, ECG: electrocardiogram, EEG-PSD: power spectral density of EEG, HRV: heart rate variability, PERCLOS: percentage of eyelid closure over the pupil over time, HRV-PSD: power spectral density of HRV, CNN: convolutional neural networks, LSTM: long short-term memory.