A Predictive Pareto Dominance Based Algorithm for Many-Objective Problems

1. Abstract Multiobjective genetic algorithms (MOGAs) have successfully been used on a wide range of real world problems. However, it is generally accepted that the performance of most state-of-the-art multiobjective genetic algorithms tend to perform poorly for problems with more than four objectives, termed many-objective problems. The contribution of this paper is a new approach for identifying members close to the true Pareto frontier. The proposed criterion, termed predicted Pareto dominance, is extendable to many-objective problems and is relatively fast to compute. Predicted Pareto dominance machine learning to generate a model of the technology frontier. The results show that, on two test problems, predicted Pareto dominance can result in improved performance of a MOGA for many-objective problems; in one of the test problems, increased performance was seen in problems with three or four objectives. The computational study also reveals at least one class of problems for which a naïve implementation of predicted Pareto dominance results in poor performance. 2.