Cellular Genetic Algorithms: Understanding the Behavior of Using Neighborhoods

Abstract

In this paper, we analyze the neighborhood effect in the selection of parents on an evolutionary algorithm. In this line, we compare a cellular genetic algorithm (cGA), which intrinsically uses the neighbor notion in the mating process, with a modified genetic algorithm including the concept of neighborhood in the selection of parents. Additionally, we analyze the neighborhood size considered for the selection of parent, trying to discover if a quasi-optimal size exists. All the analysis is carried out from a traditional analytic sense to a theoretical point of view regarding evolvability measures. The experimental results suggest that the neighbor effect is important in the performance of an evolutionary algorithm and could provide the cGA with higher chances of success in well-known optimization problems. Regarding the neighborhood size, there is an evidence that a range of neighbors of six, plus/minus two, individuals leads to the cGA to perform more efficiently than other considered sizes.