The Distance - Based Selection Technique for Crossover in Genetic Algorithm

Abstract

The aim of this research is to study and develop the natural inspired parent selections for the crossover operator in genetic algorithms. There are three distance-based methods of mating selection: the hamming distance-based selection (HS), the cosine coefficient distance-based selection (CS), and the Pearson coefficient distance-based selection (PS). The experiment conducts the comparison of the distance-based selection methods with two traditional selections: the roulette wheel selection (RWS) and the tournament selection (TS). In the experiment, all selection methods are evaluated based on four binary testing problems: one-max, zero-max, random-max, and two trap problems. [1] The measurement criterion is the number of generations when the answer is found and the fitness values when the correct answer is not found. From the experimental results, the suitable approaches are divided into two groups according to the characteristics of the benchmark problems. For the trap problem with many local optima [2], the distance-based selection methods outperformed the traditional selection. However, for the other benchmark problems, the tournament selection is the better method than others.