Table 3: Feature importances for the random forest classifiers. Impurity refers to the most important feature based on mean decrease in impurity, and permutation refers to the most important feature based on feature permutation.
| Approach | Most important feature (impurity) | Most important feature (permutation) |
|---|---|---|
| Baseline | Lactatedehydrogenase | Age |
| Layered Axes (all) | Systolic BP Day 3 Min | Cystatin C |
| Layered Axes (severe) | Serumsodium | Diastolic BP Day 3 Max |
| Layered Axes (respiratory) | Blood Sugar & Oxygen Saturation | Blood Sugar D3 Min |
| Prognosis Space (all) | Lymphocyte % | Cystatin C |
| Prognosis Space (severe) | Age | Age |
| Prognosis Space (severe, respiratory) | Total Bilirubin | Total Bilirubin |