Figure 5
From: Dynamic ensemble prediction of cognitive performance in spaceflight
A variable's importance was measured by the increase in mean squared error (MSE) by permuting that variable in a random forest model. Variable importance rankings were obtained from 100 resampling draws. The resulting "Top 10 Rate" (x-axis) describes how a given variable, over resampling trials, is repeatedly among the 10 most important variables in a random forest model. We then defined the Top 10 variables as those which scored higher on this metric; these consisted of the lagged LRM-50 through Total Sleep Missed (green text). Arrows next to names refer to the direction of association in the linear mixed effects model (Table S8): " <  < " represents association with lower LRM-50 (better performance) and vice versa for " >  > ". Due to low counts, the EVA variables were not included in the linear mixed effects model. (OPS = Overall Performance Score; RST = Reaction Self-Test; ISS = International Space Station; EVA = extravehicular activity).
