A regretbased incremental elicitation for multicriteria force design V. MakHau a , T. Cao b , J. Yearwood a a School of Information Technology, Deakin University, Waurn Ponds, Vic 3216, Australia b Joint and Operations Analysis Division, Defence Science and Technology Group, Australia Email: vicky.mak@deakin.edu.au Abstract: We aim to develop a framework for defence force design multiobjective decisionmaking problems where there are multiple decision makers each with a potentially different set of priorities, but only one solution is required in the end in practice. Benabbou et al. proposed an interactive preference elicitation approach designed for problems with a single decision maker in the Proceedings of the AAAI Conference (2020). In this approach, a linear scalarizing function is used for the multiple objectives. In each iteration, a population of different scalarizing parameter vectors are generated and the associated singleobjective optimization problems are solved one by one, each providing a different “optimal” solution. Following that, the search space of the scalarizing parameters is reduced through preference elicitation of pairwise “optimal” solutions obtained from the parameter vector population as each preference produces a cut to the feasible parameter set. The stopping criteria is controlled by the calculation of a regret value. If, in any iteration, out of all solutions found, there is a solution that returns the smallest maximum pairwise regret value and that the value is smaller than a predetermined tolerance, then the solution will be chosen as the final optimal solution. In this paper, we propose a number of adjustments and modifications to the Benabbou et al. approach required for our intended application  Defence Force Design. We provide an insight on the impact of having multiple decision makers instead of just a single decision maker. It may seem counterintuitive, but with multiple decision makers, we expect that the search space of the linear scalarizing function parameters converges to a single point much quicker if the decision makers conflict with each other in their preferences. We also propose a hierarchical approach to deal with high dimensional objective space as well as alternative scalarizing function parameter search schemes. Keywords: Multiobjective optimization, Regret model, preference elicitation, multiple decision makers 24th International Congress on Modelling and Simulation, Sydney, NSW, Australia, 5 to 10 December 2021 mssanz.org.au/modsim2021 KEYNOTE 939