TABLE V.
Accuracy of depression severity classification with and without feature selection.Voice refers to the vocal features (see Section III-B). See Section V-B for definition of features for alternative methods.
| Modality | Mean Accuracy (%) | |
|---|---|---|
|
| ||
| All features | mRMR | |
| Current Study | ||
|
| ||
| Facial Mov. Dynamics | 65.68 | 72.59 |
| Head Mov. Dynamics | 59.87 | 65.25 |
| Voice | 43.54 | 44.44 |
|
| ||
| Alternative Methods | ||
|
| ||
| Facial Mov. Dynamics [41] | 54.26 | 64.98 |
| Facial Movements [14] | 51.46 | 59.50 |
| Head Mov. Dynamics [41] | 50.68 | 56.06 |
| Head Mov. GMM [19] | 45.63 | 47.76 |
| Head Mov. Functionals [19] | 47.29 | 53.00 |
| Voice [11] | 42.33 | 42.33 |
Note: Alternative methods were re-implemented from their description in the corresponding papers for the measurement of 3-levels of depression severity. “All features” refers to no feature selection and mRMR refers to the mRMR feature selection method. Best performing approach is boldfaced for each modality/feature type.