Fig. 5

Random Forest models assessment. (A) Precision and Recall for 100 realisations of the three RF models, with their 95% confidence intervals and medians. Marker shape accounts for the model evaluated (i.e., trained using these features and subsets thereof), while colours represent the resampling method used. Line opacity stands for whether the data was subject to a PCA (dark lines) or not (transparent lines). (B) F1 score for 100 realisations of the three RF models, with medians and 95% confidence intervals. (C) Differential contribution of TML and clinical features to model's predictive performance (model 1, without PCA or resampling) for the 100 realisations, with medians and 95% confidence intervals