Figure 8
From: Dynamic ensemble prediction of cognitive performance in spaceflight
At each time point and for a given astronaut, LRM-50 can be predicted by fitting the model on all preceding data from that astronaut and the full data from other astronauts. The prediction (solid green line) is then made using covariate values from that time point. Sampling variability, given by the height of the error bars at each predicted value, was estimated as the interquartile range of predicted values from 100 bootstrap samples. In order to preserve time-varying trends, individuals were bootstrapped. The actual values of LRM-50 are displayed as hollow yellow circles. To protect astronauts' privacy, the x-axis is the proportion of elapsed mission time rather than calendar time. (A) A participant with prediction error in the highest (worst) 25th percentile (Root mean squared error (RMSE) = 8.87). (B) A participant with prediction error in the lowest (best) 25th percentile (RMSE = 4.56).
