Fig. 6: Understanding important risk factors for mortality prediction from tree-based models based on different follow-up times.
From: Interpretable machine learning prediction of all-cause mortality
a Relative importance of input features in 1-, 3-, 5- and 10-year mortality models. For each model, the figure shows the 20 most important features of prediction (ordered by importance). The purple line indicates that the feature is in the top 20 features of two models. Blue and red lines indicate that the feature is in the top 20 features of one model, but not in the top 20 features of the other. b The SHAP value of serum potassium in the 1-year mortality model. c The SHAP value of serum potassium in the 5-year mortality model. d The SHAP value of serum sodium in the 1-year mortality model. e The SHAP value of serum sodium in the 5-year mortality model.
