Figure 4
From: An AUC-based permutation variable importance measure for random forests
Distribution of AUC-values for 100 simulated datasets for AUC-based (filled) and error-rate-based (unfilled) permutation VIMs for different class imbalances. The AUC is used to assess the ability of a VIM to discriminate between noise predictors and predictors with a weak (left panel), moderate (middle panel) and strong (right panel) effect. Distributions are shown for a total sample size of n = 100.