Robust indicator-based algorithm for interactive evolutionary multiple objective optimization

We propose a novel robust indicator-based algorithm, called IEMO/I, for interactive evolutionary multiple objective optimization. During the optimization run, IEMO/I selects at regular intervals a pair of solutions from the current population to be compared by the Decision Maker. The successively provided holistic judgements are employed to divide the population into fronts of potential optimality. These fronts are, in turn, used to bias the evolutionary search toward a subset of Pareto-optimal solutions being most relevant to the Decision Maker. To ensure a fine approximation of such a subset, IEMO/I employs a hypervolume metric within a steady-state indicator-based evolutionary framework. The extensive experimental evaluation involving a number of benchmark problems confirms that IEMO/I is able to construct solutions being highly preferred by the Decision Maker after a reasonable number of interactions. We also compare IEMO/I with some selected state-of-the-art interactive evolutionary hybrids incorporating preference information in form of pairwise comparisons, proving its competitiveness.