FIGURE 3 from Integrative Prognostic Machine Learning Models in Mantle Cell Lymphoma

<p>Feature importance in the combined feature XGBoost models. <b>A,</b> The VIP for the top 20 features from the “all features” XGBoost model. B2M, beta-2 microglobulin; BM, bone marrow involvement; BMI, body mass index; dx, diagnosis; ECOG, Eastern Cooperative Oncology Group performance status; GI, gastrointestinal; Hgb, hemoglobin; LDH, lactase dehydrogenase; WBC, white blood cell count. <b>B,</b> SHAP additive values demonstrating feature impact on model output from the “all features” XGBoost model. Colors represent the value of the feature; note continuous features have a color gradient, while the dummy variables values of 0 and 1 separate into two groups “low and high”. <b>C,</b> The VIP for features that were included in the parsimonious model or iMIPI. The top 20 features were selected from the most important features from the “all features” model; 23 features are represented in the figure due to categorical variables being transformed into dummy variables during data preprocessing. <b>D,</b> SHAP values from the parsimonious model (iMIPI).</p>