Table 6.
Models with the best AUCs obtained in the experiment based on the segmentation of the ECG signal into 180 s implementing the LOSO cross-validation.
| Model | N° Features |
Hyperparameters | AUC (%) |
SEN (%) |
SPE (%) |
GDR (%) |
FDH (h−1) |
FDR (%) |
F1 (%) |
Time Delay (s) |
|---|---|---|---|---|---|---|---|---|---|---|
| (Mean ± Standard Error) | ||||||||||
| Linear SVM | Full feature set (26) | λ = 10−7
Solver: dual C1 = 3; C2 = 1 |
56 ± 5 | 22 ± 7 | 87 ± 3 | 31 ± 8 | 1 ± 0.2 | 6 ± 1 | 20 ± 6 | 141 ± 4 |
| Features selected with mRMR (2) |
λ = 10−7
Solver: dual C1 = 1; C2 = 40 |
58 ± 5 | 22 ± 9 | 77 ± 5 | 25 ± 9 | 1 ± 0.2 | 4 ± 1 | 13 ± 1 | 138 ± 15 | |
| Gaussian SVM | Full feature set (26) | Box Constraint: 0.5 Kernel Scale: 25 C1 = 1; C2 = 5 |
50 ± 4 | 51 ± 1 | 61 ± 5 | 58 ± 10 | 2 ± 0.3 | 10 ± 1 | 27 ± 6 | 117 ± 13 |
| Features selected through mRMR (2) | Box Constraint: 5 Kernel Scale: 0.1 C1 = 1; C2 = 200 |
62 ± 5 | 47 ± 8 | 67 ± 3 | 62 ± 9 | 3 ± 0.3 | 16 ± 1 | 29 ± 5 | 123 ± 3 | |