Figure 1.

Analysis of variables available for the training of the ExoGrail model through the application of the Boruta algorithm via bootstrap resampling. 1000 resamples with replacement of the available data were made, with the normalised permutation importance of each variable recorded at each iteration, along with the decision of Boruta within that resample. Fill colour shows the proportion of resamples that a feature was positively retained by Boruta. Those features selected in ≥90% of resamples were selected for fitting predictive models. Variables rejected in all of the 1000 resamples are not shown here but are fully detailed in Table S1.