Conceptdiscriminant function
- ID
- discriminant-function
- Kind
- function
- Name
- discriminant function
- Description
The discriminant function associated with a classification model. In the binary case, a discriminant function $g(x)$ for a classifier has the property that when $g(x) > 0$, the classifier predicts 1 and when $g(x) < 0$, the classifer predicts 0. A threshold other zero may also be used. For probabilistic models, the discriminant function is often just the log-odds, but this need not be the case. A notion of discriminant function may also be defined for multicategory classification. A classification model does not uniquely define a discriminant function, because composing a discriminant function with any strictly monotone function yields an equivalent decision function (although a different threshold may be needed to get the same predictions). In practice, most classification methods have a "standard" discriminant function derived from a probability model or optimization problem. The term "discriminant function" is used by some authors (see references) but is not entirely standard. The term "decision function" is also sometimes used, especially in the literature on support vector machines, but should probably be avoided to prevent confusion with the standard notion of "decision rule" from statistical decision theory.
- Inputs
- model: classification-model
- predictors: data
- Outputs
- discriminant: vector