Skip to main content
. 2022 Aug 29;19(17):10736. doi: 10.3390/ijerph191710736

Table 3.

Performance of the CNN and CNN-LSTM models for the five-level drowsiness classifications for different combinations of input signals.

Model Name Input Signal Accuracy (%) Sensitivity (%) Specificity (%) Precision (%)
CNN HRV 51 50 48 51
HRV-PSD 52 53 54 52
RR 55 59 56 52
HRV + HRV-PSD 57 56 56 52
HRV + RR 59 60 61 56
HRV-PSD + RR 57 59 52 56
All inputs 64 62 67 62
CNN-LSTM HRV 54 50 51 50
HRV-PSD 58 56 63 54
RR 60 60 65 63
HRV + HRV-PSD 61 60 61 59
HRV + RR 63 62 62 63
HRV-PSD + RR 60 55 62 53
All inputs 67 67 70 66

HRV: heart rate variability, HRV-PSD: power spectral density of HRV, RR: respiration rate.