Table 6 Comparison of the model performance on the train and validation set. For the computation of the sensitivity and specificity the threshold for each model for positive predictions was chosen such that the sensitivity on the training set is 95%
Model | Train | Validation | ||||
|---|---|---|---|---|---|---|
AUC (%) | Sensitivity (%) | Specificity (%) | AUC (%) | Sensitivity (%) | Specificity (%) | |
Conventional statistics (rule-based) | 93.7 (90.1–96.6, 95% CI) | 89.7 (82.5–95.6, 95% CI) | 97.7 (96.9–98.6, 95% CI) | 91.1 (80.6–98.9, 95% CI) | 84.6 | 97.7 (97.3–98.1, 95% C.I |
Decision tree | 97.0 (95.0–98.6, 95% CI) | 95.4 (90.7–99.7, 95% CI) | 89.1 (87.2–90.8, 95% CI) | 98.0 (96.7–99.0, 95% CI) | 100 | 87.0 (86.1–87.8, 95% CI) |
Logistic regression | 99.3 (98.8–99.7, 95% CI) | 95.4 (90.5–98.9, 95% CI) | 96.8 (95.6–97.7, 95% CI) | 98.8 (98.3–99.2, 95% CI) | 100 | 89.7 (88.8–90.5, 95% CI) |
Random forest | 99.3 (98.6–99.7, 95% CI) | 95.4 (90.3–98.9, 95% CI) | 96.9 (95.9–97.8, 95% CI) | 99.4 (98.8–99.8, 95% CI) | 100 | 95.7 (95.1–96.2, 95% CI) |
XGBoost | 99.3 (98.8–99.8, 95% CI) | 95.4 (90.6–99.0, 95% CI) | 96.8 (95.7–97.7, 95% CI) | 99.4 (98.5–99.9, 95% CI) | 100 | 93.7 (93.1–94.3, 95% CI) |