Table 5.
Performance of different methods on the Sleep-EDFx and UCDDB datasets.
| Methods | Channels | Overall Performances | |||||
|---|---|---|---|---|---|---|---|
| Reference | Feature | Classifier | Refinement | Acc | MF1 | Kappa | |
| Sleep-EDFx | |||||||
| Supratak et al. [17] | CNN & RNN | FC | - | 1 | 82.0 | 76.9 | 0.76 |
| Tsinalis et al. [36] | Handcraft | AE | - | 1 | 78.9 | 73.7 | - |
| Yu et al. [37] | ACNN | FC | - | 1 | 82.8 | 77.8 | - |
| Phan et al. [20] | ARNN | SVM | - | 1 | 82.5 | 72.0 | 0.76 |
| Mousavi et al. [38] | CNN & ARNN | FC | - | 1 | 84.3 | 79.7 | 0.79 |
| Zhang et al. [24] | MB-CNN | FC | Expert Rules | 4 | 83.6 | 78.1 | 0.77 |
| Proposed | MB-CNN | FC | FC | 4 | 85.8 | 81.2 | 0.80 |
| UCDDB | |||||||
| Shi et al. [19] | Handcraft | RF | - | 2 | 81.1 | - | - |
| Martin et al. [39] | Handcraft | DBN | HMM | 3 | 72.2 | 70.5 | 0.64 |
| Yuan et al. [40] | MB-CNN | FC | - | 4 | 74.2 | 68.2 | - |
| Cen et al. [41] | CNN | FC | HMM | 4 | 69.7 | - | - |
| Proposed | MB-CNN | FC | FC | 4 | 79.4 | 78.8 | 0.73 |
ACNN: attentional CNN, ARNN: attentional RNN, AE: auto encoder, DBN: deep brief nets.