Fig. 4: Optimal sub-model (GBDT) SHAP feature importance heatmap.

The GBDT model demonstrated the best performance among all sub-models. Feature importance was visualized using SHapley Additive exPlanation (SHAP) summary plots, revealing that the number of resuscitations (NOR), duration of mechanical ventilation (DOMV), length of stay in the ICU until discharge (LOC-ICU), diastolic blood pressure (DBP), along with other clinical features, had the most significant impact on outcomes in the GBDT model.