Table 2.
Classification models based on all PPG features.
| Classification problem | Accuracy | Interpretation features | Classification features | ||
|---|---|---|---|---|---|
| Mean | Max | Min | |||
| BSLandSDB | 0.93 | 0.95 | 0.90 | 110(SampEn), 105(LF), 109(LF/HF), 108(nLF), 107(TP), 23[Std(T)], 26[Std(T3)], 99[Rmssd(Ha)], 65[Rmssd(Rs)], 97[Rmssd(Hw)] | 110, 105, 109, 23, 26 |
| BSLandCPT | 0.98 | 0.98 | 0.98 | 99[Rmssd(Ha)] | 99 |
| BSLandMAT | 0.84 | 0.87 | 0.78 | 85[Mean(He)], 101[Rmssd(He)], 90[Std(Hz)], 108(nLF), 109(LF/HF), 99[Rmssd(Ha)], 100[Rmssd(Hb)], 70[Mean(TWst)], 93[Std(He)], 16[Mean(H1)] | 85, 90, 108, 99 |
| BSLandSDBandCPTandMAT | 0.80 | 0.84 | 0.78 | 105(LF), 94[Std(Hb/Ha)], 110(SampEn), 109(LF/HF), 108(nLF), 99[Rmssd(Ha)], 107(TP), 85[Mean(He)], 102[Rmssd(Hb/Ha)], 21[Mean(Rs)] | 105, 94,109, 99, 102, 23 |
The interpretation and classification features corresponding to the highest classification accuracy in 30 runs are listed. For those that had more than 10 interpretation features, only the top 10 are listed. Features are presented in descending order of feature importance.
BSL, baseline state; SDB, slow deep breathing; CPT, cold pressor test; MAT, mental arithmetic test; Mean, mean accuracy of 30 runs at different random states; Max, maximum accuracy; Min, minimum accuracy; Std, standard deviations; Rmssd, root mean square of the successive difference; SampEn, sample entropy of pulse rate; LF, low-frequency power of PRV; TP, total power of PRV; nLF, normalized low-frequency power; LF/HF, ratio of LF to HF; T, time span of the pulse; T3, diastolic time of the pulse; TWst, stress index; H1, height of the systolic peak; RS, rising slope of PPG; Hw, height of w point in VPG (first derivative or velocity of PPG); Hz, height of z point in VPG; Ha, height of a point in APG (second derivative or acceleration of PPG); Hb, height of b point in APG; He, height of e point in APG.