a The AUC averaged over five folds is 0.82 ± 0.02. b Predicted probability distribution for mild/moderate versus severe/critical cases (area normalized to 1). The class imbalance influences the classifier probabilities. c Reliability plot done by bootstrapping the logistic regression with different 80–20% train-test splits repeated 20,000 times. The predicted probability is more accurate below 25% which is where most samples are located. d Normalized distribution of the bootstrapped predictive probability.
展开▼