A feature’s relative importance in determining whether a simulation will follow the Type II evolutionary scenario. The feature variables are described in Table 3. The error bars represent the standard deviation for the feature importance as calculated from 50 different random forest models trained on the same data. Note that even within error, ﹩{M}_{m}^{{{\rm{H}}}_{2}{\rm{O}}}﹩ (initial water in the mantle) is by far the most important variable, with ﹩{\rm{\Delta }}{E}_{{{\rm{H}}}_{2}{\rm{O}}}﹩ (the coupling strength between mantle water content and viscosity) next.