By Jordi de la Torre on September 30, 2015
Model ensembling is a way to improve accuracy predictions using a combination a set of classifiers that are built using different models, data, hyperparameters or parameter initializations. Every difference introduced in the ensembling increases diversity and improves generalization capabilities of the final model.
The drawbacks are the loss of interpretability and the increase in complexity of the evaluation of the ensembled model.
It takes a set (lots of) weak predictors, it weights them and add them up. The combination of them is a stronger predictor.
It randomizes the selection of input variables to create a set of individual classification/regression trees and it combines the predictions of all of them.