Generalization refers to a model’s ability to perform well on previously unseen data. Models that overfit have poor generalization, whereas those that underfit don’t capture enough detail. Balancing these extremes is crucial for robust real-world performance.