Overfit

Part of speech: verb

Definitions

1. To excessively adapt a model to its training data, leading to diminished predictive ability on new instances
2. The process of customizing a predictive algorithm too finely to specific data points, adversely affecting performance on unfamiliar data sets
3. The act of fine-tuning a statistical model to the point where it accurately reflects the training data but fails to generalize to unseen examples

Etymology: "Overfit" is a plain compound: the prefix "over-", meaning excessively or beyond the proper amount, attached to "fit" in the sense of adjusting something so it matches a requirement. To fit a garment or a part is to shape it to the measure needed, so to overfit is to shape something past that point, making it too closely tailored to one set of measurements and ill-suited to others. The verb's modern sense comes from statistics and machine learning, where a model that memorizes the quirks and noise of its training data performs well on that data but poorly on new examples. The older everyday pattern of "over-" plus a verb, as in "overcook" or "overeat," gave the technical term its shape, so the sense of excess carries over directly into the data-science meaning.

Synonyms: overtrain

Antonyms: generalize