Concept - Generalization

Definition (s.331): In ML a model is used to fit the data; once trained, it is applied to new data. Generalization is the prediction capability of the model on live/new data.

• s.332 “Which model is better?”: a straight line vs a wiggly S-shaped curve separating SPAM from Not SPAM points. The simple line makes a few training mistakes; the wiggly curve fits all training points but is fitted to noise and is not expected to generalize.
• A model that generalizes poorly is Concept - Overfitting (complex model, s.333); too simple a model gives Concept - Underfitting.

Related

Concept - Model Complexity | Concept - Overfitting | Concept - Underfitting | Concept - Bias-Variance Tradeoff

Source chunk: 03 Text Classification - Chunk 12 (s.331-360)