TABLE 2.
Performance metrics of machine‐learning models in predicting mortality with a threshold of 0.5 on the validation cohort
| Model | Accuracy (95% CI) | Sensitivity | Specificity | PPV | NPV | Kappa | F1 | Brier |
|---|---|---|---|---|---|---|---|---|
| XGBoost | 0.719 (0.686–0.751) | 0.720 | 0.718 | 0.756 | 0.679 | 0.436 | 0.738 | 0.275 |
| RF | 0.723 (0.690–0.755) | 0.782 | 0.651 | 0.732 | 0.711 | 0.437 | 0.756 | 0.271 |
| LRLasso | 0.596 (0.560–0.631) | 0.309 | 0.945 | 0.872 | 0.529 | 0.237 | 0.456 | 0.396 |
| SVM | 0.690 (0.656–0.723) | 0.727 | 0.645 | 0.714 | 0.661 | 0.373 | 0.720 | 0.304 |
| KNN | 0.702 (0.668–0.734) | 0.656 | 0.759 | 0.767 | 0.644 | 0.408 | 0.707 | 0.292 |
Abbreviations: 95% CI, 95% confidence interval; KNN, K‐nearest neighbor; LRLasso, logistic regression with lasso regularization; NPV, negative predictive value; PPV, positive predictive value; RF, random forest; SVM, support vector machine; XGBoost, extreme gradient boost.