Bootstrap sampling repeatedly draws training observations with replacement, creating diverse subsets for multiple models. Repeated observations may appear, while others remain out of bag. section 5.3.2
Within bagging and Random Forests, each decision tree trains on a bootstrap subset. Averaging or voting across trees reduces variance and smooths individual tree errors. section 5.3.2
Bootstrap sampling supports:
- Diverse decision trees trained from different versions of the dataset. section 5.3.2
- Out-of-bag evaluation using observations excluded from each tree’s training subset, avoiding a separate validation set. section 5.3.2
CYP inhibition classification
A RandomForestClassifier trains 200 trees with out-of-bag scoring on molecular fingerprints for cytochrome P450 inhibition classification, reporting validation and OOB metrics.






