Supervised Evolution: Research Concerning the Number of Evolutions that Occur Under Certain Constraints

Abstract

It is known that evolution may lead to a new species while adaptation may lead to a new variety. In this manuscript, we present an analysis of the number of evolutions (defined by improvement of the score associated with an objective function of a genetic algorithm) in an experiment supervised by a genetic algorithm, experiment conducted on octan-1-ol/H20 partition coefficient of polychlorinated biphenyls. The numbers of evolutions resulted from 9 implemented evolution strategies were investigated. Evolutions arisen from the first 20 000 generations coming from 46 independent runs were recorded. A distribution analysis has been conducted for each evolution strategy. Without exception, the Weibull distribution fits well with the number of evolutions at a significance level of 5% for any evolution strategy. Furthermore, the Weibull distribution could not be rejected when different merged samples were investigated.