Figure 2.
Top global model explanations from ML models to predict severe chest injury. X-axis represents rescaled variable importance from (1) EN beta coefficients, (2) XGB permuted feature importance, and (3) CNN training dataset averaged LIME explanations. Y-axis represents the preferred text definition of CUIs; CUI codes omitted here for clarity. CNN LIME interpretation median r2 = 0.69 (IQR 0.46–0.93).