A study of crossover operators in genetic programming

Abstract

Holland's analysis of the sources of power of genetic algorithms has served as guidance for the applications of genetic algorithms for more than 15 years. The technique of applying a recombination operator (crossover) to a population of individuals is a key to that power. Neverless, there have been a number of contradictory results concerning crossover operators with respect to overall performance. Recently, for example, genetic algorithms were used to design neural network modules and their control circuits. In these studies, a genetic algorithm without crossover outperformed a genetic algorithm with crossover. This report re-examines these studies, and concludes that the results were caused by a small population size. New results are presented that illustrate the effectiveness of crossover when the population size is larger. From a performance view, the results indicate that better neural networks can be evolved in a shorter time if the genetic algorithm uses crossover.