How does feature engineering influence bias and variance?

Feature engineering can reduce bias by adding informative signals, but may increase variance if it introduces noisy or overly specific features.

Feature engineering directly impacts model capacity and generalization. Adding meaningful transformations (e.g., log transforms, interactions, domain-specific encodings) helps reduce bias by making patterns easier to learn. However, adding too many irrelevant or high-dimensional features increases variance because the model can start fitting noise. In high-dimensional spaces, even simple models can overfit due to sparsity.