What is bias-variance decomposition in ensemble models?

Bias-variance decomposition breaks prediction error into bias, variance, and irreducible noise components.

In ensemble learning, bias measures systematic error, variance measures sensitivity to training data, and noise is inherent data randomness. Bagging reduces variance by averaging unstable models, while boosting reduces bias by iteratively correcting errors. Ensemble methods aim to optimize this decomposition by combining models with complementary error profiles.