Table 2.
Model performance (in percentage) and elements of confusion matrix of the various stacked learners in EMPaSchiz model: average (standard errors) − 5 × 10-fold CV
| Accuracy | Precision | Sensitivity | Specificity | True positive | True negative | False positive | False negative | |
|---|---|---|---|---|---|---|---|---|
| Stacked-multi | 86.9 (1.1) | 91.9 (1.4) | 79.8 (1.8) | 93.1 (1.2) | 65.0 (1.4) | 86.8 (1.2) | 6.2 (1.1) | 16.0 (1.4) |
| Stacked-ALFF | 76.4 (1.4) | 76.3 (1.8) | 73.9 (2.2) | 78.7 (1.9) | 59.8 (1.7) | 73.0 (1.7) | 20.0 (1.9) | 21.2 (1.8) |
| Stacked-ReHo | 74.1 (1.6) | 73.4 (2.0) | 74.6 (2.0) | 73.6 (2.5) | 60.4 (1.6) | 68.2 (2.3) | 24.8 (2.5) | 20.6 (1.6) |
| Stacked-fALFF | 74.5 (1.5) | 73.8 (1.7) | 72.2 (1.8) | 76.6 (1.9) | 58.6 (1.6) | 72.0 (1.7) | 21.0 (1.7) | 22.4 (1.7) |
| Stacked-FC-correlation | 82.4 (1.3) | 83.9 (1.9) | 79.7 (1.8) | 84.7 (2.0) | 64.6 (1.5) | 78.8 (2.0) | 14.2 (1.9) | 16.4 (1.4) |
| Stacked-FC-partial correlation | 78.5 (1.4) | 93.7 (1.5) | 58.2 (2.8) | 96.2 (0.9) | 46.8 (2.4) | 89.8 (1.0) | 3.2 (0.8) | 34.2 (2.3) |
| Stacked-FC-precision | 83.7 (1.2) | 90.2 (1.6) | 73.8 (2.0) | 92.3 (1.3) | 60.0 (1.9) | 86.8 (1.3) | 6.2 (1.2) | 21.0 (1.8) |
| Baselinea | 51.2 (0.3) | 47.0 (0.5) | 40.7 (0.6) | 60.2 (0.5) | 33.0 (0.4) | 56.0 (0.5) | 37.0 (0.5) | 48.0 (0.5) |
aBaseline results are based on permutation test over the randomly shuffled labels (based on 100 repetitions of entire ‘learning with subsequent 10-fold CV evaluations’)