Table 4 Comparison of feature importances calculated by random forest and SHAP. SHAP, Shapley additive explanations; CRP, C-reactive protein; IL6, interleukin-6; PCT, procalcitonin; WBC, white blood cell count; ADV, adenovirus; FLUA, influenza virus A; RSV, respiratory syncytial virus; PIV, Human parainfluenza viruses. Random forest importance, feature importance as determined by the random forest model, averaged across all classes. SHAP importance, feature importance as determined by SHAP values, averaged across all classes.
From: Using random forest and biomarkers for differentiating COVID-19 and Mycoplasma pneumoniae infections
Feature | Random forest | SHAP |
|---|---|---|
Age | 0.25 | 0.27 |
CRP | 0.21 | 0.25 |
IL6 | 0.19 | 0.14 |
PCT | 0.12 | 0.14 |
WBC | 0.12 | 0.05 |
ADV (Yes) | 0.04 | 0.07 |
Sex (Male) | 0.03 | 0.02 |
FLUA (Yes) | 0.02 | 0.04 |
RSV (Yes) | 0.01 | 0.01 |
PIV (Yes) | 0.01 | 0.01 |