Fig. 3: Combining 5-year mortality prediction gradient boosted trees models and local explanations to achieve significant discoveries about the entire model and individual features.
From: Interpretable machine learning prediction of all-cause mortality
a SHAP summary plot for the gradient boosted trees trained on the 5-year mortality prediction task. The plot shows the most impactful features on prediction (ranked from most to least important) and the distribution of the impacts of each feature on model output, which includes a set of plots where each dot corresponds to an individual. The colors represent feature values for numeric features: red for larger values, and blue for smaller. The thickness of the line comprised of individual dots is determined by the number of examples at a given value. A negative SHAP value (extending to the left) indicates reduced mortality risk, while a positive one (extending to the right) indicates increased mortality risk. b, c IMPACT can verify well-studied features associated with mortality. b The main effect of red cell distribution width on 5-year mortality. c The main effect of serum albumin on 5-year mortality. d–h IMPACT can identify less well-studied features associated with mortality. d The SHAP value for arm circumference in 5-year mortality model. e The main effect of platelet count on 5-year mortality. f The main effect of serum chloride on 5-year mortality. g The SHAP interaction value of serum chloride vs. age in the 5-year mortality model. h The SHAP interaction value of serum chloride vs. gender in the 5-year mortality model.
