Evaluating and Optimizing Metaheuristic Algorithms:A step ahead in VRP research

Abstract

The evaluation of heuristic performance on vehicle routing problems (VRP) is most commonly done through empirical testing on some benchmark problem set. Parameters are set using trial-and-error and testing on a limited set of standard test problems. This makes it difficult to estimate the robustness of proposed algorithms. An algorithm working well on a set of standard test problems does not necessarily perform well on any particular data set and is usually extremely sensitive to changes in the data. Therefore a scientific basis is needed for heuristic evaluation and comparison, and is the subject of this research project.