TABLE III.
Accuracy of depression severity classification using different per-frame and per-video based encoding for facial movement dynamics and head movement dynamics.
| Representation | Classification Accuracy (%) | Weighted Kappa | ||||
|---|---|---|---|---|---|---|
|
|
|
|||||
| Frame | Video | Remission | Mild | Moderate to Severe | Mean | |
| Facial Movement Dynamics | ||||||
|
| ||||||
| SDAE Outputs | IFV | 67.57 | 65.71 | 84.48 | 72.59 | 0.62±0.059 |
| SDAE Outputs | DFS | 64.86 | 60.00 | 79.31 | 68.06 | 0.53±0.062 |
| Raw | IFV | 59.46 | 60.00 | 81.03 | 66.83 | 0.52±0.062 |
| Raw | DFS | 56.76 | 57.14 | 81.03 | 64.98 | 0.50±0.063 |
|
| ||||||
| Head Movement Dynamics | ||||||
|
| ||||||
| SDAE Outputs | IFV | 62.16 | 54.29 | 79.31 | 65.25 | 0.51 ±0.063 |
| SDAE Outputs | DFS | 59.46 | 48.57 | 77.59 | 61.87 | 0.48±0.064 |
| Raw | IFV | 56.76 | 45.71 | 75.86 | 59.44 | 0.46±0.064 |
| Raw | DFS | 54.05 | 40.00 | 74.14 | 56.06 | 0.40±0.065 |
Note: Weighted kappa is the proportion of ordinal agreement above what would be expected to occur by chance.