Scientific evaluation of (meta)heuristic algorithms: a longitudinalanalysis approach.

Optimization problems abound in a wide variety of business environments, ranging from logistics and supply chain management, to the financial sector and health care. Companies in these areas support many of their decisions on powerful optimization methods. By doing so, they are able to become not only more profitable and competitive, but also more environmentally friendly. Metaheuristics are considered to be the dominant approach to tackle most optimization problems found in practice. However, they completely lack a scientifically sound methodology to analyze the optimization they carry out. All existing methodologies are based on the flawed principle of an algorithmic race: metaheuristics are evaluated at a single and arbitrary point in time only. Optimization is, however, not a race but a complex process that deserves a thorough assessment during its entire execution. It turns out that an analysis with these considerations has been carried out for over 40 years in clinical research. This approach, called longitudinal analysis, is commonly used to evaluate trials for determining the effectiveness of new treatments (in which several individuals are examined multiple times). The main goal of this proposal is to develop a sound methodology, based on the rigorous approach of longitudinal analysis, to evaluate (meta)heuristics. The methodology will be implemented as an open source software package, to allow researchers to apply it in a straightforward and userfriendly way.

Keywords:METAHEURISTICS

Disciplines:Applied mathematics in specific fields, Statistics and numerical methods