Fig 1.
Definitions of mortality prediction for patients with hypothermia.
Fig 2.
Overall process of mortality prediction in patients with hypothermia.
Fig 3.
Process for selection of the study cohort.
Table 1.
Non-invasive physiological parameters included in the study.
Fig 4.
Partitioning into training and testing sets.
Table 2.
Metrics used to evaluate the prediction performance.
Table 3.
Baseline characteristics of included patients.
Fig 5.
Trends in AUC values for different observation windows and prediction windows for (A) Random Forest, (B) Logistic Regression, (C) XGBoost, and (D) Naïve Bayes four machine learning methods.
Observation window length is set to 1h, 2h, 3h and 4h, respectively; Pre_n: prediction window with length n hours.
Fig 6.
Trends for average AUC for (A) different prediction window and (B) different observation window for each model.
Table 4.
Identification results of four algorithms on test sets for different feature subsets.
Fig 7.
Feature selection results for (A) Random Forest, (B) Logistic Regression, (C) XGBoost, and (D) Naïve Bayes.
The central line represents the mean, while the gray area denotes one standard error range. Triangles and dots, along with their associated numbers, signify the MIN_subset and OPT_subset, respectively. A) The BER of random forest, B) The BER of logistic regression, C) The BER of XGBoost, D) The BER of Naive Bayes.
Fig 8.
Proportion of Feature Importance.