Table 2.
Classification performance of the optimal model including different input features using the integrated machine learning framework (tenfold).
| Input feature | Feature Selection Method | Classifier | Accuracy (%) | Sensitivity (%) | Specificity (%) | AUC | p valuea |
|---|---|---|---|---|---|---|---|
| Gut microbiota features (n = 77) | RFE | RF | 70.8 | 58.3 | 83.3 | 0.80 | 0.03 |
| Blood features (n = 12) | Noneb | KNN | 83.3 | 83.3 | 83.3 | 0.88 | 0.010 |
| EEG features (n = 574) | RFE | RF | 79.2 | 83.3 | 75.0 | 0.90 | 0.010 |
| Combined features (n = 663) | None | SVM | 91.7 | 91.7 | 91.7 | 0.97 | 0.010 |
AUC area under the receiver operating characteristic curve, RFE recursive feature elimination, KNN k-nearest neighbor, LR logistic regression, RF random forest, SVM support vector machine, EEG electroencephalogram.
aThe statistical significance of the permutation test was set to p < 0.05.
bNone means no feature selection algorithm was used.