Figure 3
From: Predicting phenotypes of asthma and eczema with machine learning
Feature importance evaluation by means of random forests. Importance is calculated and shown as the rescaled mean (standard deviation) decrease in accuracy over 1000 independent runs (green colour). Boxplots represent a null feature importance distribution obtained by permuting the outcome randomly for 1000 times. Variables significant at the 0.1 level (p-values in red) are shown for current asthma (upper panel), current wheeze (middle panel), and doctor's diagnosed eczema (lower panel) using the whole feature set as input.