The ability for a machine learning model to generate predictions on new data after it has been trained. Generalization represents the ability for a given model to perform well, in general, on new data for a specific task that the model has been trained on.