Phenomenon in ML where the prediction error on test data initially decreases, increases, and then decreases again as model complexity grows.
Generality: 715
