Figure 4
From: Predicting the complexity and mortality of polytrauma patients with machine learning models
The performance metrics of the complexity prediction models for polytrauma patients upon ICU admission. (A) The bar chart of the mean and standard deviation of the performance metrics (Accuracy, Recall, F-score, AUC) for predicting the complexity of polytrauma patients upon ICU admission using four machine learning models (SVM, RF, XGBoost, ANN) in the discovery cohort was shown. (B) The performance indicators of the optimal model for predicting ICU admission complexity in the validation cohort was compared against the commonly used scores, including ISS, TI, GCS. (C) The top 15 features that contribute to the predictive model of the complexity of polytrauma patients upon ICU admission were shown. FBF, Facial bone fracture; BL, Blood loss; FBEJ, Fracture below elbow joint; APVI, Abdominopelvic visceral injury; SBP, Systolic blood pressure; UA, Uric Acid; SG, Specific Gravity; HR, Heart Rate; TRF, Thoracic rib fracture; ICH, Intracranial hematoma; EOS%, Eosinophil percentage; MCHC, Mean corpuscular hemoglobin concentration; AST, Aspartate aminotransferase; Fracture above elbow joint (FAEJ).
