Figure 2.

Predictive performance of fifteen ML algorithms on training data, as measured by the Area Under the Receiver Operating Characteristic Curve (AUC) when using 1 to 27 variables as predictors in the machine learning algorithms. Figure 2a–d demonstrate the predictive performance for 3-day, 10-day, 30-day and 365-days mortality, respectively. Among all models, the highest predictive performances in AUC are shown between 0.90 and 0.93 in Figure 2a–d. When using five variables, the top 3 models achieved an AUC of 0.89 in Fig. 2a–c, and an AUC of 0.86 or above when using five variables in Fig. 2d. The AUC falls below 0.85 when using fewer than three variables for all models in Fig. 2a–d.