Table 2.
Detailed classification metrics for all evaluated model families under strict subject-wise 10-fold cross-validation (Protocol B).
| Model family | Accuracy (%) | ROC–AUC | Precision | Recall (sensitivity) | Specificity | F1-Score |
|---|---|---|---|---|---|---|
| Logistic Regression | 59.9 ± 4.7 | 0.63 ± 0.04 | 0.641 | 0.640 | 0.548 | 0.641 |
| Random Forest | 65.1 ± 6.2 | 0.70 ± 0.04 | 0.694 | 0.670 | 0.627 | 0.682 |
| Support Vector Machine (PSD) | 65.6 ± 4.9 | 0.72 ± 0.03 | 0.696 | 0.680 | 0.625 | 0.688 |
| LSTM (Stand-alone) | 61.6 ± 7.7 | 0.63 ± 0.05 | 0.639 | 0.720 | 0.485 | 0.677 |
| CNN–BiLSTM (Hybrid) | 65.7 ± 6.8 | 0.74 ± 0.03 | 0.694 | 0.691 | 0.614 | 0.692 |
| CNN (Stand-alone) | 72.7 ± 9.3 | 0.72 ± 0.04 | 0.705 | 0.880 | 0.534 | 0.783 |
Accuracy and ROC–AUC are reported as mean ± SD across the ten folds; precision, recall, specificity, and F1-score are pooled across the aggregated test predictions (N = 4, 173). Pooled metrics are computed across all concatenated out-of-fold epoch predictions; subjects with more usable epochs contribute proportionally more to pooled scores. Bold row indicates the overall best-performing architecture under subject-wise evaluation across all metrics.