Interactively Learning the Preferences of a Decision Maker in Multi-objective Optimization Utilizing Belief-rules

Many real life problems can be modelled as multiobjective optimization problems. Such problems often consist of multiple conflicting objectives to be optimized simultaneously. Multiple optimal solutions exist to these problems, and a single solution cannot be said to be the best without preferences given by a domain expert. Preferences can be used to find satisfying solutions: optimal solutions, which best match the expert’s preferences. To model the preferences of the expert, and aid him/her in finding satisfying solutions, a novel method is proposed. The method utilizes machine learning combined with belief-rule based systems to adaptively train a belief rule based system to learn a domain expert’s preferences using preference information gathered during an interactive process. Belief-rule based systems are explainable generalized expert systems, which have not been used before in the manner described in this paper to model preferences of a domain expert for a multi-objective optimization problem. In the case study conducted, the satisfying solutions found using learned preferences are concluded to be compatible with the preferences of the expert, which support the proposed method’s viability as a decision making support tool.