18.5 Global Explanations

• Which features are most important in driving the predictions aggregated over the whole training set
• Measure how much does a model’s performance change if the effect of a selected explanatory variable(s) is(are) removed
  • If variables are correlated, then models like random forest are expected to spread importance across many variables
  • Dependent on the random nature of the permutations
  

boost_vip <- model_parts(explainer_boost, loss_function = loss_root_mean_square)

png(file="images/18_boost_vip.png")
plot(boost_vip, max_featuers = 10)
dev.off()