Contrasting main selection methods in genetic algorithms

Abstract

In genetic algorithms selection mechanisms aim to favour reproduction of better individuals imposing a direction on the search process. It does not create new individuals; instead it selects comparatively good individuals from a population and typically does it according to their fitness. The idea is that interacting with other individuals (competition), those with higher fitness have a higher probability to be selected for mating. In that manner, because the fitness of an individual gives a measure of its "goodness", selection introduces the influence of the fitness function to the evolutionary process. Moreover, selection is the only operator of genetic algorithm where the fitness of an individual affects the evolution process. In such a process two important, strongly related, issues exist: selective pressure and population diversity. They are the sides of the same coin: exploitation of information gathered so far versus exploration of the searching space. Selection plays an important role here because strong selective pressure can lead to premature convergence and weak selective pressure can make the search ineffective [14]. Focussing on this equilibrium problem significant research has been done. In this work we introduce the main properties of selection, the usual selection mechanisms and finally show the effect of applying proportional, ranking and tournament selection to a set of well known multimodal testing functions on simple genetic algorithms. These are the most widely used selection mechanisms and each of them has their own features. A description of each method, experiment and statistical analyses of results under different parameter settings are reported.