Analysis of selection, mutation and recombination in genetic algorithms

Abstract

Genetic algorithms have been applied fairly successful to a number of optimization problems. Nevertheless, a common theory why and when they work is still missing. In this paper a theory is outlined which is based on the science of plant and animal breeding. A central part of the theory is the response to selection equation and the concept of heritability. A fundamental theorem states that the heritability is equal to the regression coefficient of parent to offspring. The theory is applied to analyze selection, mutation and recombination. The results are used in the Breeder Genetic Algorithm whose performance is shown to be superior to other genetic algorithms.