What is the ability of a machine learning model to perform well on unseen data after training called?

The ability of a machine learning model to perform well on unseen data after training is referred to as generalization. This concept is central to machine learning because it indicates how effectively a model can apply what it has learned from the training dataset to new, previously unseen examples.

Generalization is crucial for the successful deployment of machine learning models, as the ultimate goal is not merely to perform well on the training data, but to make accurate predictions on new data that may not resemble the training set. If a model generalizes well, it can confidently make inferences or predictions based on its training, providing valuable insights or outputs in real-world applications.

In contrast, if a model has not generalized well and is overfitting, it has learned the training data too closely, including noise and outliers, and may fail to perform adequately on new data. Concepts like accuracy and model complexity relate to specific aspects of model performance or structure but do not directly define the ability to generalize to unseen data.