Fig. 3.

Solid lines indicate ROC for the validation set (n = 35), where the model was trained on a separate set. Dashed lines indicate miRNA models trained using a leave-one-out cross validation approach across the whole data set. (a) extreme gradient boosting (XGBoost) utilising 8 miRNAs; (b) LASSO utilising 13 miRNAs; (c) regression partition trees (Rpart) utilising 4 miRNAs; (d) a random forest wrapper method (Boruta) utilising 10 miRNAs; (e) Ensemble approach utilising 20 miRNAs (f) Average cross validated ROC for miRNA-187-5p and miRNA-636 on the training set.