Verification of performance of Multi-Parental Real-Valued Crossover in Interactive Genetic Algorithm

In interactive evolutionary computation, it is important to reduce user fatigue and to enhance the search performance in a few generations, since fatigue has an impact on users' evaluation process. These issues also have been found in in interactive genetic algorithm (IGA). In this study, we adopt multi-parental unimodal normal distribution crossover (UNDX-
$m$
) to solve these problems in IGA. UNDX-
$m$
generates many offspring with considering the distribution of selected parents. The next population can be generated by applying UNDX-
$m$
once with more than two parent solutions. Through this mechanism, the evaluation process becomes simple manner where users only select their preferable solutions. In addition, UNDX-
$m$
can generate offspring that inherit their parents' traits, which enhance the search quality. In this paper, we evaluate the effectiveness of UNDX-
$m$
through the numerical experiment, by using a color combination problem in Web design.