Figure 1
From: Selection of important variables by statistical learning in genome-wide association analysis
Rank of risk SNPs in random forest as noise level increases. The five risk SNPs (τ1-τ5) for CAC were tested with 3 environment factors and different numbers of noise SNPs (see text). At each level of noises, the test was repeated 100 times. We define relative rank of a variable to be the rank of variable importance normalized by the total number of predictors. Lower value indicates the variable is easier to be detected by random forest. The plot shows the quantiles of relative rank for the five risk SNPs. The 3 kinds of curves represent 50th, 30th and 10th quantiles, respectively (marked as "50%", "30%" and "10%" in the plot).