Table 6.
Accuracy comparison between state-of-the-art approaches on JAFFE, CK, and FER2013.
| Dataset | Validation scheme | Methods | Accuracy (%) |
|---|---|---|---|
| JAFEE | 10-fold | KDIsomap [74] | 81.6 |
| EDL [75] | 90.3 | ||
| CBIN [76] | 86.7 | ||
| Proposed approach | 92.4 | ||
| Split (70–30%) | RAU's [46] | 86.3 | |
| CNN [77] | 76.5 | ||
| Proposed approach | 96.8 | ||
|
| |||
| CK | 10-fold | DTAN + DTGN [31] | 91.4 |
| DNN [22] | 90.9 | ||
| Proposed approach | 82.8 | ||
| Split (70–30%) | DCNN as SCAE [78] | 92.5 | |
| Proposed approach | 86.5 | ||
|
| |||
| FER2013 | Split (70–30%) | HOG + C4.5 [79] | 46.1 |
| CNN [80] | 57.1 | ||
| CNN + decision tree [79] | 58.8 | ||
| VGG-Face + FR-Net-A + B + C + Audio [81] | 60.0 | ||
| AlexNet [82] | 61.0 | ||
| CNN ensemble [83] | 62.4 | ||
| Proposed approach | 62.7 | ||