Skip to main content
. 2022 Feb 22;3(4):100452. doi: 10.1016/j.patter.2022.100452

Figure 4.

Figure 4

Visual summary of MIMIC study results from SHAP-ShapleyVIC framework, part II: ShapleyVIC analysis of nearly optimal models

(A) ShapleyVIC suggested a different variable ranking after accounting for the variability in variable importance across the 350 nearly optimal models.

(B) Distribution of variable importance (indicated by the shape of violin plots) and the corresponding model performance (indicated by color) to complement inference on average ShapleyVIC values. Dark blue strips towards the right end of a violin plot suggests the presence of good models that relied heavily on the variable (e.g., creatinine) for further investigations.