Table 4.
Diagnostic performance achieved in the test set by each feature selected and each MLP optimized model for the binary classification in each severity threshold. Results are shown in terms of sensitivity (Se %), specificity (Sp %), accuracy (Acc %) and AUC.
| Threshold: AHI = 1 e/h | ||||
| Feature/model | Se | Sp | Acc | AUC |
| VLF_f2m | 44.5 | 72.3 | 53.9 | 0.605 |
| LF_BE2 | 42.1 | 72.7 | 52.4 | 0.581 |
| HF_PE | 42.9 | 63.3 | 49.8 | 0.55 |
| BW2_RPDiag | 50.9 | 64.8 | 55.6 | 0.629 |
| BW2_BE1 | 47.1 | 59.4 | 51.3 | 0.559 |
| BWRes_Bmin | 40.5 | 57.4 | 46.2 | 0.482 |
| BWRes_BE3 | 41.5 | 57.4 | 46.9 | 0.513 |
| MLP1Classic | 52.3 | 59.4 | 54.7 | 0.6 |
| MLP1Specific | 76.3 | 38.3 | 63.4 | 0.627 |
| Threshold: AHI = 5 e/h | ||||
| Feature/Model | Se | Sp | Acc | AUC |
| VLF_f2m | 62.5 | 72.2 | 70.8 | 0.749 |
| LF_BE2 | 56.3 | 74.4 | 71.7 | 0.67 |
| HF_PE | 45.5 | 72.1 | 68.2 | 0.628 |
| BW2_RPDiag | 60.7 | 77.7 | 75.2 | 0.747 |
| BW2_BE1 | 56.3 | 70.1 | 68 | 0.671 |
| BWRes_Bmin | 58.9 | 45.3 | 47.3 | 0.567 |
| BWRes_BE3 | 47.3 | 58.4 | 56.8 | 0.569 |
| MLP5Classic | 50.9 | 86.2 | 81 | 0.774 |
| MLP5Specific | 62.5 | 84.2 | 81 | 0.791 |
| Threshold: AHI = 10 e/h | ||||
| Feature/Model | Se | Sp | Acc | AUC |
| VLF_f2m | 63.8 | 76.7 | 75.6 | 0.784 |
| LF_BE2 | 58 | 81.5 | 79.4 | 0.74 |
| HF_PE | 53.6 | 72.1 | 70.4 | 0.663 |
| BW2_RPDiag | 68.1 | 76 | 75.3 | 0.789 |
VLF: very low frequency; LF: low frequency; HF: high frequency; MLP: multi-layer perceptron; AHI: Apnea–Hypopnea Index. The highest ACC and AUC for each severity threshold are highlighted in bold.