Do bagging and boosting can be used with logistic regression?
Classic MLLevel: IntermediateML concepts — Sebastian Raschka FAQ
Do bagging and boosting can be used with logistic regression?
Hints
- Separate the linear score from the function mapping it to a probability.
- Connect the target distribution to its negative log-likelihood.
Answer
I am not sure if bagging would make much sense for logistic regression -- in bagging, we reduce the variance of the deep decision tree models that overfit the training data, which wouldn't really apply to logistic regression.
Boosting could work though, however, I think that "stacking" would be a better approach here. Stacking would be more "powerful" since we don't use a pre-specified equation to adjust the weight, rather, we train a meta-classifier to learn the optimal weights to combine the models.
Here's one of the many interesting, related papers, I recommend you to check out :)
- "Is Combining Classifiers with Stacking Better than Selecting the Best One?"
Source and licence: Sebastian Raschka — Python Machine Learning FAQ · MIT