An ensemble learning strategy that sequentially trains a series of weak models, each one attempting to correctly predict the observations that the previous model got wrong.
As opposed to bagging, boosting trains on all the data and combines weak models using the learning rate .