EVALUATING EVOLUTIONARY MULTI-OBJECTIVE OPTIMIZATION ALGORITHMS USING RUNNING PERFORMANCE METRICS
With the popularity of evolutionary multi-objective optimization (EMO) methods among researchers and practitioners, an increasing interest has grown in developing new and computationally efficient algorithms and in comparing them with existing methods. Unlike in single-objective optimization in which often the goal is to find a single optimal solution, an EMO method attempts to find a set of well-converged and well-distributed set of trade-off solutions. In comparing two or more EMO methods, it is intuitive that more than one performance metrics are necessary. Although there exist a number of performance metrics in the EMO literature, they are usually applied to the final non-dominated set obtained by an EMO algorithm to evaluate its performance. In this chapter, we emphasize the need of running performance metrics, which will provide the dynamics of the working of an EMO algorithm. Either using a known Pareto-optimal front or an agglomeration of generation-wise populations, two suggested metrics reveal important insights and interesting dynamics of the working of an EMO and help provide a comparative evaluation of two or more EMO methods.