Figure 4.

(A), (B), (C), and (D) Sum of variable importance values for all variables across 10 random runs, by model: Gradient Boosting (A), Random Forest (B), ANN (C), and SVM (D). All models besides ANN consistently chose NPI as the most important variable. Other important variables include tumor size and stage, ER/PR/HER2 status, and breast surgery status. K-means cluster with the worst survival was moderately important across models except for ANN. ANN was the most unstable model in terms of the values of variable importance assigned to each variable across runs. The x-axis denotes the sum of variable importance values across 10 random runs and may not exceed 1000, which is the sum for a variable that was the most important through all runs of a model.