Fig. 2: Variable importance measurement from RF by MDA for the total dataset (n = 38) with 5000 trees grown and 100 random permutations.
From: Machine learning classification of ADHD and HC by multimodal serotonergic data
Predictors are ordered by declining information criterion for classification accuracy. Blue lines each represent drops in MDA after a random permutation of the predictor values and the red line shows averages of all 100 permutations. The precentages indicate how often the top predictors for the whole dataset were selected in the training sets of the cross-validation runs and thus were used for classification of the test sets. a shows results for SNPs while b shows ROI predictors. RF RandomForest, MDA mean decrease in accuracy, SNP single nucleotide polymorphisms, ROI region of interest.
