Title :
Directing crossover for reduction of bloat in GP
Author :
Terrio, M.D. ; Heywood, M.I.
Author_Institution :
Fac. of Comput. Sci., Dalhousie Univ., Halifax, NS, Canada
Abstract :
A method is proposed to reduce the amount of invisible code (or bloat) produced in individuals while searching for a parsimonious solution under tree structured genetic programming. Known as directed crossover this process involves the identification of highly fit nodes to use as crossover points during operator application. Three test problems, including medical data classification, are used to assess the performance of directed crossover when applied at various thresholds. Results, collected over 1260 independent runs, identify conditions under which directed crossover reduces code bloat.
Keywords :
genetic algorithms; genetics; liver; medical diagnostic computing; trees (mathematics); benchmark medical data sets; breast liver data sets; chromosome crossover; code bloat; directed crossover; highly fit nodes; independent runs; individuals; introns; invisible code; medical data classification; operator application; parsimonious solution; steady state tournament selection; test problems; thresholds; tree structured genetic programming; Bioinformatics; Biological cells; Computer science; Genetic algorithms; Genetic programming; Genomics; Machine learning; Medical tests; Network-on-a-chip; Resource management;
Conference_Titel :
Electrical and Computer Engineering, 2002. IEEE CCECE 2002. Canadian Conference on
Print_ISBN :
0-7803-7514-9
DOI :
10.1109/CCECE.2002.1013102
