Discovering Design Principles for Soft Multi-objective Decision Making

Abstract

Decision-making tasks sometimes involve soft objectives. They are soft in the sense that they contain uncertainty, imprecision or vagueness. For example, decisions on built environment aim to maximize comfort or other experiential qualities. Pareto-optimal solutions to such problems can be found using multi-objective evolutionary search together with other soft computing methods. Beyond optimality, professionals are interested in knowing how different aspects of the problem influence each other in optimal solutions. Such knowledge is referred to as the design principles. Through them decisions can be taken with great confidence and knowledge for similar future design cases is gained. Using the Pareto-optimal solutions for this purpose is known as innovization and it has been exercised for various crisp engineering problems. In the present paper automated innovization is used to discover the principles for a soft decision making problem. The process involves the use of a grid-based clustering technique integrated with a genetic algorithm for unsupervised learning of the principles. Multiple design principles are discovered simultaneously through a niching strategy. The large number of variables originating from the softness of the problem poses an additional challenge of parsimonious knowledge representation for ensuring interpretability. The problem investigated is a real-world decision making task concerning the optimal placement of a number of residential units in an urban design, involving two soft objectives: the recuperative quality of the neighborhood, as well as its living comfort should both be maximized. The underlying design principles are obtained and interpreted from the point of view of the decision-maker. This demonstrates the relevance of evolutionary knowledge discovery in decision-making, a matter which should provide decision-makers adequate and informed knowledge for choosing a single preferred solution among the Pareto-optimal ones, and also to understand intricate trade-offs among decision variables.