Clasificación mediante programación genética

  1. J. Riquelme 1
  2. F. Fernández Bejarano 1
  3. P. González Morón 1
  4. M. Toro 1
  1. 1 Universidad de Sevilla
    info

    Universidad de Sevilla

    Sevilla, España

    ROR https://ror.org/03yxnpp24

Livre:
CAEPIA'97: actas
  1. Botti, Vicent (coord.)

Éditorial: Vicent Botti ; Asociación Española para la Inteligencia Artificial (AEPIA)

ISBN: 978-84-8498-765-9 84-8498-765-5

Année de publication: 1997

Pages: 571-580

Congreso: Conferencia de la Asociación Española para la Inteligencia Artificial. (7. 1997. null)

Type: Communication dans un congrès

Résumé

A method to obtain a classifier is presented in this paper. The initial information is a real space n-dimensional with two sets of points of different characteristics. The aim is to define in a symbolic way the codimension 1 surface that separate each set. We use a parametric regression that tries to approximate the analytic formula of an unknown function from the coordinates of a set of points and their values for this function. In order to carry out the parametric regression, the use of Genetic Programming (GP) is proposed. The GP is a variation of Genetic Algorithms where the individuals are trees. The use of GP for regression was proposed in (Koza 92a) and has as evident advantage on the statistic regression, that it have to know previously the form of the function. The function Φ found can be used as classifier for the space, as positive values of Φ would indicate a characteristic, and the negative values the opposite one. This technique has been applied in several examples and the results have been very satisfactory.