Tournament selection based on statistical test in genetic programming

Selection plays a critical role in the performance of evolutionary algorithms. Tournament selection is often considered the most popular techniques among several selection methods. Standard tournament selection randomly selects several individuals from the population and the individual with the best fitness value is chosen as the winner. In the context of Genetic Programming, this approach ignores the error value on the fitness cases of the problem emphasising relative fitness quality rather than detailed quantitative comparison. Subsequently, potentially useful information from the error vector may be lost. In this paper, we introduce the use of a statistical test into selection that utilizes information from the individual’s error vector. Two variants of tournament selection are proposed, and tested on Genetic Programming for symbolic regression problems. On the benchmark problems examined we observe a benefit of the proposed methods in reducing code growth and generalisation error. © Springer International Publishing AG 2016.