The Distributional Biases of Crossover Operators

The choice of genetic operators is one way in which genetic algorithms can be tailored to speciic optimization problems. For bit represented problems, the choice of crossover operator , or the choice not to use a crossover operator, can dramatically aaect search performance. The eecacy of crossover for genetic search is governed by the relationship between the crossover biases and the search problem itself. Crossover operators have two forms of bias: positional bias and distribu-tional bias. This paper analytically characterizes the distributional biases that exist for several commonly used crossover operators. The eeects of the crossover biases are empirically examined for a Simple Genetic Algorithm applied to two types of NK-landscapes.