Bootstrap sampling creates a training sample by drawing observations from an original dataset with replacement. For a dataset of N observations, each sample contains N draws; observations may appear more than once, while others may be omitted. section 9.2.1 section 9.3.3
In bagging, each sample trains a model, and the models’ predictions are aggregated. Random forests use a new bootstrap sample for each tree, alongside random feature selection. Resampling creates variation, but does not guarantee that trees will differ enough to prevent correlated errors. section 9.2.1 section 9.2.3 section 9.3.3






