Figure 2. Feature importance for the 10 most important input variables for each of the 4 outcomes based on the gradient boosting classifier models.

Feature importance, a measure inherent to tree‐based algorithms, is higher the more a variable contributes to the prediction of a specific outcome. Accordingly, the NIHSS score on admission was the most important variable in prediction of in‐hospital mortality, increased intracranial pressure, and deep vein thrombosis, whereas administration of thrombolytic therapy was the most telling variable in the prediction of an intracerebral hemorrhage. Individual feature importance has been normalized by the top‐ranked input variable and therefore range from 0 to 1. NIHSS indicates National Institutes of Health Stroke Scale.