Addressing Evolutionary-Based Dynamic Problems: A New Methodology for Evaluating Immigrants Strategies in MOGAs

Abstract

Multi-Objective Genetic Algorithms (MOGAs) have been successfully used to address dynamic problems in a wide variety of domains. In these domains, data changes over time, so a non-static analysis is required to obtain feasible solutions. In this type of environments, MOGAs are often time-consuming and require special adaptation to work properly. A number of different techniques have been proposed to adapt MOGAs to dynamic environments for tackling the previous problems such as hypermutation, memory and immigrant schemes or multi-population methods, among others. In particular, immigrant strategies are one of the most commonly used methods, for that reason, this work proposes a new methodology that allows to make a detailed evaluation of their performance when these strategies are used. The proposed methodology works on two levels, a coarse-grain one and a fine-grain one. In the former, an overall evaluation of the different immigrant strategies is made based on three different dimensions: Quality, Stability and Speed. In the latter, a detailed study of the status of the immigrant individuals during the evolution of the algorithm is carried out. This is a very relevant aspect to take into account in order to evaluate whether an immigrant strategy is working properly or not. To deploy this methodology, a new visualization technique for population mixing analysis is proposed in this work. In order to validate the proposed methodology, a test case in the context of the Dynamic Community Detection problem (DCD) has been selected using a MOGA that applies several different immigrant schemes, showing both how the methodology works and how it could be employed in a particular dynamic problem.