Dataset-level explanations

Permutation-based Importance

Below are permutation-based importance measures originally introduced in chapter 16. Here we use the default 10 permutations per model.

Partial Dependence Profiles

Sample R Code for Feature Importance and PDP Profiles

Permutation-based feature importance:

fifa_mp_gbm_deep <- model_parts(fifa_gbm_exp_deep)
plot(fifa_mp_gbm_deep, max_vars = 15, 
     bar_width = 4, show_boxplots = FALSE) 

PDP Profile:

selected_variables <- c("movement_reactions", "skill_ball_control", "skill_dribbling", "age")
fifa19_pd_deep <- model_profile(fifa_gbm_exp_deep, 
                             variables = selected_variables)
plot(fifa19_pd_deep)