Bagging, short for bootstrap aggregating, is training different base learners on different subsets of the training set randomly, by drawing random training sets from the given sample (with replacement).