Table 3.
Machine-learning results for both EP and TG datasets.
| Classifiers | Protocol | Accuracy (95% CI) | Sensitivity (95% CI) | Specificity (95% CI) |
|---|---|---|---|---|
| Support Vector Machine Linear (L-SVM) |
1 | 0.808 (0.751–0.862) | 0.752 (0.623–0.860) | 0.863 (0.803–0.918) |
| 2 | 0.795 (0.739–0.851) | 0.670 (0.578–0.762) | 0.920 (0.867–0.973) | |
| 3 | 0.818 (0.776–0.861) | 0.771 (0.705–0.836) | 0.866 (0.813–0.919) | |
| Support Vector Machine 3rd order Polynomial (P-SVM) |
1 | 0.774 (0.708–0.835) | 0.813 (0.687–0.921) | 0.736 (0.645–0.820) |
| 2 | 0.780 (0.723–0.837) | 0.670 (0.578–0.762) | 0.890 (0.829–0.951) | |
| 3 | 0.729 (0.680–0.778) | 0.873 (0.820–0.925) | 0.586 (0.509–0.663) | |
| Support Vector Machine Radial basis function (R-SVM) |
1 | 0.811 (0.750–0.870) | 0.755 (0.621–0.861) | 0.867 (0.805–0.923) |
| 2 | 0.795 (0.739–0.851) | 0.640 (0.546–0.734) | 0.950 (0.907–0.993) | |
| 3 | 0.815 (0.772–0.858) | 0.752 (0.684–0.819) | 0.879 (0.828–0.930) | |
| Decision Tree | 1 | 0.761 (0.706–0.812) | 0.733 (0.624–0.826) | 0.789 (0.728–0.849) |
| 2 | 0.625 (0.558–0.692) | 0.290 (0.238–0.422) | 0.960 (0.953–1.000) | |
| 3 | 0.764 (0.717–0.811) | 0.796 (0.733–0.859) | 0.732 (0.663–0.802) | |
| Random Forest | 1 | 0.815 (0.754–0.869) | 0.789 (0.662–0.900) | 0.842 (0.784–0.896) |
| 2 | 0.655 (0.589–0.721) | 0.330 (0.238–0.422) | 0.980 (0.953–1.000) | |
| 3 | 0.809 (0.765–0.852) | 0.796 (0.733–0.859) | 0.822 (0.762–0.882) | |
| Multi-layer Perceptron (MLP) | 1 | 0.796 (0.733–0.855) | 0.759 (0.617–0.873) | 0.833 (0.760–0.899) |
| 2 | 0.800 (0.745–0.855) | 0.660 (0.567–0.753) | 0.940 (0.893–0.987) | |
| 3 | 0.701 (0.650–0.751) | 0.904 (0.858–0.950) | 0.497 (0.419–0.575) | |
| Logistic Regression | 1 | 0.813 (0.757–0.869) | 0.754 (0.602–0.880) | 0.873 (0.816–0.926) |
| 2 | 0.800 (0.745–0.855) | 0.670 (0.578–0.762) | 0.930 (0.880–0.980) | |
| 3 | 0.831 (0.790–0.873) | 0.866 (0.813–0.919) | 0.796 (0.733–0.859) | |
| K-nearest Neighbors (KNN) | 1 | 0.780 (0.719–0.833) | 0.799 (0.678–0.894) | 0.760 (0.689–0.825) |
| 2 | 0.795 (0.739–0.851) | 0.670 (0.578–0.762) | 0.920 (0.880–0.980) | |
| 3 | 0.806 (0.762–0.849) | 0.803 (0.740–0.865) | 0.809 (0.747–0.870) |
Protocol 1: Both TG and EP dataset with leave-one-subject-cross-validation, Protocol 2: EP for training and TG for testing, Protocol 3: TG for training and EP for testing. CI: confidence interval.