MOSES: A Metaheuristic Optimization Software EcoSystem

Abstract

Many problems that we face nowadays can be expressed as optimization problems. Finding the best solution for real-world instances of such problems is hard or even infeasible. Metaheuristic algorithms have been used for decades to guide the search for satisfactory solutions in hard optimization problems at an affordable cost. However, despite its many benefits, the application of metaheuristics requires overcoming numerous obstacles. First, the implementation of efficient metaheuristic programs is a complex and error-prone process. Second, since there is no analytical method to choose a suitable metaheuristic program for a given problem, experiments must be performed. Besides this, experiments are usually performed ad-hoc, with generic tools and no clear guidelines, introducing threats to validity, and making them hard to automate and reproduce. Our aim is to reduce the cost of applying metaheuristics for solving optimization problems. To that purpose, a set of tools to support the selection, configuration and evaluation of metaheuristic-based applications is presented.