ABC, a new performance tool for algorithms solving dynamic optimization problems

Measuring the performance is still an important unsolved issue in dynamic optimization. Although several measures have been proposed in the literature, the problem about which ones should be used to describe the behaviour of algorithms remains open. One of the aspects to be considered is whether fitness averages are able to summarize the overall performance of metaheuristics over dynamic problems. Another issue is how to compare algorithms and, more specifically, how to quantify the numerical difference in performance between them. The main goal in this article is to propose a new way of measuring the behaviour of algorithms and also to provide a method to quantify the distance between them. We introduce thus two measures: one based on the area below the curve defined by some population property at each generation (e.g., the best-of-generation fitness), and a second one based on the area between the curves of two different algorithms.