On the Search Properties of Different Crossover Operators in Genetic Programming

In this paper we study and compare the search properties of different crossover operators in genetic programming (GP) using probabilistic models and experiments to assess the amount of genetic material exchanged between the parents to generate the offspring. These operators are: standard crossover, one-point crossover and a new operator, uniform crossover. Our analysis suggests that standard crossover is a local and biased search operator not ideal to explore the search space of programs effectively. One-point crossover is better in some cases as it is able to perform a global search at the beginning of a run, but it suffers from the same problems as standard crossover later on. Uniform crossover largely overcomes these limitations as it is global and less biased