Fig. 5

Performance of random forest models predicting generalism (binary classification, present at more than one area or not). “All” means a blocked cross-validation with each phylum as a fold (Actinobacteria: red squares, Bacteroidetes: green triangles, Firmicutes: blue diamonds, Proteobacteria: purple circles). Within each phylum we performed blocked cross-validation using classes as folds, except in the case of Bacteroidetes where all species in the dataset were in the same class and order, so that families were used as the folds. Shown are two measures of performance (\(\kappa\) and area under the precision-recall curve), as well as the prevalence of specialist species in a fold (\(P(\text {Specialist})\) for “probability is a specialist”)