Out-of-bag (OOB) evaluation estimates a Random Forest’s performance without a separate validation set. Bootstrap sampling leaves some training examples out of each tree’s sample; those excluded examples evaluate that tree, and the ensemble aggregates the resulting evaluations. section 5.3.2
This makes OOB estimates useful when data is limited and supports hyperparameter tuning or comparisons between ensemble configurations. In a CYP inhibition classifier, OOB accuracy is 0.801 and OOB PR AUC is 0.806. section 5.3.2


