Preference-Based Evolutionary Multi-Objective Optimization Using Non-Preference Information

This paper presents a novel approach to preference-based evolutionary multi-objective optimization (PBEMO) using non-preference information. Multi-objective optimization problems are common in numerous fields, such as engineering, economics, and industry, and they often involve trade-off relationships. A key challenge with these problems is identifying and incorporating preferences, which are essential for guiding satisfactory options for decision makers (DMs). Traditional methods have focused on explicit preferences provided by DMs through repetitive interactions. To avoid the inevitable frustration surrounding this task, this study introduces an innovative technique that also utilizes non-preference information (i.e., data corresponding to implied but non-selected options) based on user intentions and extant constraints to provide superior results while reducing the number of interactions required during the optimization process. It enables the discovery of solutions that more quickly approximate the DM's intentions by effectively exploiting non-preference information. A use case based on a menu-based nutrient and preference decision process demonstrates the performance of the proposed method. From this effort, a significant PBEMO gap is reduced by integrating a circumstantially more intuitive understanding of user preferences, both expressed and implied.