Table 10.
Stress recognition accuracy, sensitivity, and specificity on the constructed database.
| Method | Feature Dimension | Accuracy (%) | Sensitivity (%) | Specificity (%) |
|---|---|---|---|---|
| HOG-SVM [68,69] | 1764 | 50.9153 | 50.4360 | 64.3488 |
| VGG-16 [65] | 2048 | 56.9125 | 56.4093 | 71.0178 |
| CBAM-ResNet-18 [47] | 512 | 58.8559 | 58.1435 | 72.4161 |
| ResNet-50 [57] | 2048 | 60.0093 | 59.4649 | 74.2789 |
| ResNet-18 [57] | 512 | 60.0895 | 59.4573 | 74.4877 |
| Inception v3 [66] | 2048 | 63.4185 | 62.8578 | 77.6015 |
| AlexNet [64] | 4096 | 64.1588 | 63.4871 | 78.3340 |
| DenseNet-121 [67] | 1024 | 64.9408 | 64.4349 | 78.2179 |
| SE-ResNet-18 [46] | 512 | 65.7013 | 65.1206 | 79.2945 |
| 2D-CNN + LSTM + Facial Landmark [42] | 768 | 58.3432 | 57.7521 | 72.5148 |
| 3D-CNN + Facial Landmark Image [41] | 4096 | 62.5361 | 62.1877 | 76.1710 |
| 2D-CNN + GRU + Multimodel [43] | 512 | 65.8770 | 65.3907 | 79.3543 |
| 3D-CNN + Hyperparameter Optimize [45] | 4096 | 65.9372 | 65.4369 | 79.7895 |
| Zhang et al. [30] | 47,104 | 64.6481 | 64.0199 | 78.3209 |
| Ours (w/o Facial Landmark Feature) | 512 | 65.3396 | 64.5928 | 78.9639 |
| Ours | 768 | 66.8409 | 66.1292 | 80.0959 |