 |
    |
On Naïve Crossover Biases with Reproduction for Simple Solutions to Classification Problems
M. David Terrio and Malcolm I. Heywood
Dalhousie University, Faculty of Computer Science 6040 University Avenue, Halifax, NS. B3H 1W5 Canadamterrio@cs.dal.ca
mheywood@cs.dal.ca
Abstract. A series of simple biases to the selection of crossover points in treestructured genetic programming are investigated with respect to the provision of parsimonious solutions. Such a set of biases has a minimal computational overhead as they are based on information already used to estimate the fitness of individuals. Reductions to code bloat are demonstrated for the real world classification problems investigated. Moreover, bloated solutions provided by a uniform crossover operator often appear to defeat the application of MAPLETM simplification heuristics.
LNCS 3103, p. 678 ff.
lncs@springer.de
© Springer-Verlag Berlin Heidelberg 2004