Table 5.
Results of the LOSO evaluation for different classifiers with data of patients with tremor.
| Data Representation | # of Features | Classifier | Sensitivity | Specificity | AUC |
|---|---|---|---|---|---|
| FFT (baseline) | 64 | SVM | 81.1% (5.89) | 80.5% (6.67) | 0.872 (0.045) |
| FFT | 64 | AdaBoost (100 estimators) | 84.3% (3.20) | 84.4% (3.42) | 0.918 (0.025) |
| MFCCs [49] | 36 | AdaBoost (100 estimators) | 82.6% (4.86) | 82.2% (4.65) | 0.905 (0.040) |
| Mahadevan et al. [55] | 64 | Random Forest (10 estimators) | 82.3% (5.88) | 84.0% (4.51) | 0.893 (0.049) |
| Hssayeni et al. [15] | 39 | Gradient Boost (170 estimators) | 84.1% (4.52) | 84.0% (4.13) | 0.922 (0.035) |
| Raw signal | 384 | CNN (with GAP) | 84.1% (4.30) | 84.0% (4.20) | 0.915 (0.043) |
| Raw signal | 384 | multitask CNN | 85.0% (5.15) 2 | 85.3% (5.13) 2 | 0.923 (0.039) 2 |
| FFT | 64 | multitask CNN | 86.1% (5.37) 2 | 86.1% (5.49) 2 | 0.936 (0.024) 2 |
2 Values obtained only from the tremor detector output.