Table 6 The average performance of the proposed hybrid deep learning model on the BioVid database (Part A) in leave-one-subject-out cross-validation for classifying No Pain versus Very Severe Pain.
Modality | Model | Accuracy (%) | Sensitivity (%) | Specificity (%) | F1-score (%) | F1 95% \({\textbf {CI}}^{\dagger }\) |
|---|---|---|---|---|---|---|
ECG | FCN | 62.27 \(\pm \,13\) | 39.65 \(\pm \,32\) | 84.88 \(\pm \,20\) | 51.24 \(\pm \,27\) | [40.71, 52.69] |
ALSTM | 63.59 \(\pm \,12\) | 43.33 \(\pm \,29\) | 83.85 \(\pm \,16\) | 54.34 \(\pm \,25\) | [45.73, 56.42] | |
FCN-ALSTM-Transformer | 65.17 \(\pm \,12\) | 44.48 \(\pm \,29\) | 86.39 \(\pm \,15\) | 55.47 \(\pm \,23\) \(^*\) | [50.01, 60.21] | |
EDA | FCN | 80.05 \(\pm \,15\) | 79.13 \(\pm \,27\) | 80.97 \(\pm \,23\) | 79.87 \(\pm \,22\) | [72.84, 82.22] |
ALSTM | 84.56 \(\pm \,13\) | 78.21 \(\pm \,25\) | 90.92\(\pm \,9\) | 83.52 \(\pm \,20\) | [76.81, 85.47] | |
FCN-ALSTM-Transformer | 86.12 \(\pm \,13\) | 84.36 \(\pm \,21\) | 87.87 \(\pm \,17\) | 85.87 \(\pm \,16\) \(^{**}\) | [81.10, 88.34] | |
EDA & ECG | Crossmod-Transformer | 87.52\(\pm \,11\) | 84.59\(\pm \,17\) | 89.86 \(\pm \,13\) | 87.15\(\pm \,13\) \(^{***}\) | [84.26, 90.06] |
