Table 2 Performance of the prediction models using all features (without SMOTE).
Model | AUROC | AUPRC | Optimal threshold | Accuracy | F1 Score |
|---|---|---|---|---|---|
Internal test | |||||
Logistic regression | 0.878 (0.812, 0.934) | 0.752 (0.624, 0.863) | 0.299 | 0.844 (0.792, 0.902) | 0.722 (0.610, 0.818) |
Random forest | 0.911 (0.855, 0.956) | 0.823 (0.718, 0.905) | 0.350 | 0.861 (0.809, 0.908) | 0.745 (0.632, 0.836) |
LightGBM | 0.885 (0.826, 0.937) | 0.786 (0.662, 0.888) | 0.176 | 0.850 (0.798, 0.896) | 0.717 (0.609, 0.817) |
XGBoost | 0.870 (0.803, 0.926) | 0.722 (0.584, 0.859) | 0.084 | 0.798 (0.734, 0.855) | 0.673 (0.554, 0.764) |
External test | |||||
Logistic regression | 0.836 (0.798, 0.869) | 0.723 (0.666, 0.779) | 0.285 | 0.763 (0.731, 0.793) | 0.624 (0.570, 0.674) |
Random Forest | 0.857 (0.826, 0.889) | 0.746 (0.689, 0.792) | 0.570 | 0.828 (0.802, 0.857) | 0.577 (0.502, 0.641) |
LightGBM | 0.841 (0.805, 0.876) | 0.717 (0.657, 0.773) | 0.973 | 0.789 (0.759, 0.818) | 0.360 (0.278, 0.440) |
XGBoost | 0.847 (0.814, 0.880) | 0.731 (0.671, 0.787) | 0.945 | 0.809 (0.780, 0.837) | 0.459 (0.374, 0.531) |
