Unlike Bagging which primarily reduces variance, Boosting primarily reduces which component of error, and what is a key risk?

Question

Answer

Bias, because Boosting sequentially reduces training error, but risks overfitting if too many rounds are used or base learners are too complex.

Answer

Variance, because Boosting averages predictions from many models, similar to Bagging.

Answer

Irreducible error, because Boosting identifies and removes noisy training examples through re-weighting.

Answer

Both bias and variance equally, making it always superior to Bagging.

Crowdly