What is feature bagging in ensemble learning and why is it effective?

Feature bagging trains multiple models on random subsets of features to increase diversity and reduce overfitting.

Feature bagging is an extension of bootstrap aggregation where diversity is introduced by randomly sampling feature subsets instead of or in addition to data samples. Each base model sees a different projection of the feature space, forcing them to learn different patterns. This reduces correlation among models, which is critical because ensemble gains depend more on error independence than individual model strength. It is especially powerful in high-dimensional datasets where many features are redundant or noisy.