Table 4.
Best predictive performance results in fivefold cross-validation of classifiers trained on the simplified set and the full set of features
| Sensitivity [%] (recall) | Specificity [%] | Accuracy [%] | G-mean [%] | AUC | ||
|---|---|---|---|---|---|---|
| Logistic regression | 78 ± 25 | 30 ± 31 | 65 ± 10 | 48.4a | 54 ± 3 | Significant lesion |
| Xgboost | 56 ± 18 | 58 ± 20 | 57 ± 8 | 57.0a | 57 ± 2 | |
| DRSA-BRE (full set of features) | 56.9 ± 0.2 | 66.9 ± 0.2 | 59.6 ± 0.2 | 61.7 ± 0.02 | 61.9a | |
| Logistic regression | 47 ± 34 | 90 ± 11 | 89 ± 10 | 65.0a | 68 ± 11 | In-hospital death |
| Xgboost | 80 ± 9 | 79 ± 4 | 80 ± 4 | 79.5a | 78 ± 3 | |
| DRSA-BRE | 79.3 ± 1.7 | 80.6 ± 0.5 | 81.0 ± 0.5 | 79.9 ± 1 | 80.8a | |
| DRSA-BRE (full set of features) | 81.0 ± 2.4 | 81.1 ± 0.5 | 81.0 ± 0.5 | 81.0 ± 1 | 81.0a |
aIndicates that value was not directly estimated during experiments