What kind of problem, circumstances and data makes it more suitable to apply boosting instead of bagging methods?
$\begingroup$
$\endgroup$
1
-
$\begingroup$ If the questions been answered please mark it as so (helps to keep the unanswered questions list low) $\endgroup$– Tasty213Commented Aug 27, 2019 at 15:19
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
1
Bagging and boosting are two methods of implementing ensemble models.
Bagging: each model is given the same inputs as every other and they all produce a model
Boosting: the first model trains on the training data and then checks which observations it struggled most with, it passes this info to the next algorithm which assigns greater weight to the misclassified data
Because of this both bagging and boosting reduce variance. However boosting is better at improving accuracy vs a single model whilst bagging is better at reducing overfitting.
I would advise training a single version of the ensemble aka decision tree for random forest and then seeing where improvements can be made. Good article explaining boosting vs bagging
-
2$\begingroup$ I don’t think this mis fully answers the question (more about circumstances and data/problem than the inner working of bagging/boosting). Also, bagging does not imply the same inputs (see random forests) and boosting is more generally a method of finding a solution to an additive model which aims at correcting errors iteratively (but with different heuristics). $\endgroup$– ElliotCommented Aug 27, 2019 at 22:56