Table 4.
Summary of k-fold cross-validation for different classification models trained using the selected features of linear, nonlinear, and frequency indices of EHG.
| Classification models | Accuracy (%) | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | AUC | F-Score | ||
|---|---|---|---|---|---|---|---|---|---|
| S1F1 | Quadratic Supportive Vector Machine (SVM) | Train | 95.56 ± 0.25 | 94.72 ± 0.36 | 96.36 ± 0.42 | 96.16 ± 0.42 | 95.01 ± 0.32 | 0.94 ± 0.03 | 0.93 ± 0.03 |
| Validation | 93.52 ± 3.24 | 92.97 ± 6.09 | 94.13 ± 5.64 | 93.99 ± 5.64 | 93.44 ± 5.37 | 0.94 ± 0.03 | 0.93 ± 0.03 | ||
| Test | 82.04 ± 1.11 | 64.09 ± 2.21 | 100.00 ± 0.00 | 100.00 ± 0.00 | 73.59 ± 1.19 | 0.82 ± 0.01 | 0.78 ± 0.02 | ||
| S1F2 | Cubic SVM | Train | 97.56 ± 0.29 | 97.21 ± 0.34 | 97.89 ± 0.35 | 97.79 ± 0.35 | 97.34 ± 0.32 | 0.93 ± 0.04 | 0.93 ± 0.05 |
| Validation | 93.17 ± 4.25 | 92.56 ± 5.99 | 93.76 ± 5.33 | 93.51 ± 5.72 | 93.08 ± 5.69 | 0.93 ± 0.04 | 0.93 ± 0.05 | ||
| Test | 55.04 ± 3.31 | 25.83 ± 3.76 | 84.26 ± 4.95 | 62.69 ± 9.33 | 53.17 ± 2.04 | 0.55 ± 0.03 | 0.36 ± 0.05 | ||
| S1F3 | Fine Gaussian SVM | Train | 94.90 ± 0.170 | 94.63 ± 0.26 | 95.16 ± 0.31 | 94.94 ± 0.3 | 94.86 ± 0.22 | 0.92 ± 0.03 | 0.91 ± 0.04 |
| Validation | 91.84 ± 2.75 | 91.22 ± 6.55 | 92.07 ± 4.70 | 91.90 ± 3.98 | 92.05 ± 4.55 | 0.92 ± 0.03 | 0.91 ± 0.04 | ||
| Test | 81.13 ± 1.39 | 94.52 ± 1.24 | 67.74 ± 2.03 | 74.57 ± 1.30 | 92.52 ± 1.69 | 0.81 ± 0.01 | 0.83 ± 0.01 | ||
| S2F1 | Fine Gaussian SVM | Train | 96.63 ± 0.20 | 97.76 ± 0.25 | 95.55 ± 0.30 | 95.47 ± 0.29 | 97.80 ± 0.23 | 0.88 ± 0.04 | 0.88 ± 0.04 |
| Validation | 88.20 ± 4.28 | 88.88 ± 5.85 | 88.04 ± 6.69 | 87.75 ± 6.95 | 88.99 ± 6.09 | 0.88 ± 0.04 | 0.88 ± 0.04 | ||
| Test | 87.52 ± 1.20 | 75.04 ± 2.40 | 100.00 ± 0.00 | 100.00 ± 0.00 | 80.06 ± 1.54 | 0.88 ± 0.01 | 0.86 ± 0.02 | ||
| S2F2 | Quadratic SVM | Train | 91.70 ± 0.21 | 92.59 ± 0.38 | 90.84 ± 0.84 | 90.66 ± 0.33 | 92.74 ± 0.34 | 0.9 ± 0.04 | 0.9 ± 0.04 |
| Validation | 90.33 ± 4.23 | 91.44 ± 5.84 | 89.41 ± 6.15 | 89.47 ± 5.40 | 91.07 ± 6.05 | 0.90 ± 0.04 | 0.90 ± 0.40 | ||
| Test | 66.83 ± 1.03 | 37.57 ± 1.47 | 96.09 ± 2.05 | 90.82 ± 4.41 | 60.61 ± 0.61 | 0.67 ± 0.01 | 0.53 ± 0.01 | ||
| S2F3 | Weighted Kernel Nearest Neighbors (KNN) | Train | 74.31 ± 0.44 | 70.34 ± 4.88 | 78.12 ± 4.62 | 75.82 ± 2.99 | 73.44 ± 1.97 | 0.68 ± 0.07 | 0.66 ± 0.08 |
| Validation | 68.32 ± 3.490 | 64.16 ± 10.13 | 72.80 ± 9.32 | 69.28 ± 9.52 | 68.00 ± 9.51 | 0.68 ± 0.07 | 0.66 ± 0.08 | ||
| Test | 51.04 ± 2.98 | 40.26 ± 5.16 | 61.83 ± 9.91 | 52.07 ± 4.65 | 50.65 ± 2.32 | 0.51 ± 0.01 | 0.45 ± 0.03 | ||
| S3F1 | Bagged Tree | Train | 99.89 ± 0.09 | 99.93 ± 0.12 | 99.85 ± 0.16 | 99.84 ± 0.17 | 99.93 ± 0.12 | 0.71 ± 0.06 | 0.7 ± 0.09 |
| Validation | 71.52 ± 6.2 | 68.9 ± 11.69 | 73.53 ± 9.78 | 71.35 ± 9.62 | 72.08 ± 5.98 | 0.71 ± 0.03 | 0.70 ± 0.09 | ||
| Test | 47.91 ± 3.360 | 32.70 ± 3.60 | 63.13 ± 50 | 47.1 ± 4.61 | 48.36 ± 2.65 | 0.48 ± 0.03 | 0.39 ± 0.04 | ||
| S3F2 | Cubic SVM | Train | 99.10 ± 0.22 | 99.3 ± 0.27 | 98.9 ± 0.29 | 98.86 ± 0.30 | 99.33 ± 0.12 | 0.93 ± 0.03 | 0.92 ± 0.04 |
| Validation | 92.64 ± 3.18 | 93.16 ± 5.26 | 92.26 ± 4.56 | 91.73 ± 5.92 | 93.28 ± 5.02 | 0.93 ± 0.03 | 0.92 ± 0.04 | ||
| Test | 72.17 ± 2.25 | 84.43 ± 1.59 | 59.91 ± 4.11 | 67.87 ± 2.30 | 79.35 ± 2.09 | 0.72 ± 0.02 | 0.75 ± 0.02 | ||
| S3F3 | Quadratic SVM | Train | 93.46 ± 0.31 | 93.57 ± 0.53 | 93.35 ± 0.44 | 93.11 ± 0.41 | 93.8 ± 0.48 | 0.91 ± 0.04 | 0.91 ± 0.04 |
| Validation | 91.30 ± 3.65 | 92.10 ± 5.86 | 90.33 ± 5.14 | 90.37 ± 4.55 | 92.52 ± 5.31 | 0.91 ± 0.04 | 0.91 ± 0.04 | ||
| Test | 88.52 ± 1.47 | 83.83 ± 3.07 | 93.22 ± 1.31 | 92.51 ± 1.46 | 85.27 ± 2.49 | 0.89 ± 0.02 | 0.88 ± 0.02 |
The dataset employed for each classifier is denoted as S (1,2,3) for the signal channel and the frequency subband with F (1,2,3; 0.3 – 1 Hz, 1 – 2 Hz, 2 – 3 Hz, respectively). The highest classification accuracy is shown in bold.