Using Distribution-based Operators in Genetic Algorithms
Rafael Nogueras, Carlos Cotta · 2005
Genetic Algorithms (GAs) constitute a very efficient search model that has provided excellent results in different domains during the last fifty years. However, new methods offering additional possibilities are emerging. Estimation of distribution Algorithms (EDAs) are one of these methods. In this work, we study the combination of both approaches. To be precise, we consider the use of a Bayesian Network (BNs) to improve the best individuals found by the GA. BNs are a probabilistic model that we utilize to predict the performance of a particular individual in its subsequent mutations and crossovers. This can be used to provide hints on which the most convenient way to mutate one particular individual, or to recombine two different individuals is.