Large network analysis for fisheries management using coevolutionary genetic algorithms

Traditionally, a genetic algorithm is used to analyze networks by maximizing the modularity (Q) measure to create a favorable community. A coevolutionary algorithm is used here to not only find the appropriate community division for a network, but to find interesting networks containing substantial changes in data within a very large network space. The network is one of the largest, if not the largest, analyzed by evolutionary computation techniques to date and is created using a real world data set consisting of fisheries catch data in the north Atlantic Ocean off the coast of Canada. This work examines the quantitative performance of two types of coevolutionary algorithms against both a standard GA that uses a natural (but not necessarily optimal) division of the data set into communities, and simulated annealing. The goal for all search algorithms was to automatically find anomalies (differences in catch) within the data. To measure practical usefulness of the system, a fisheries expert analyzed the best networks located by the search algorithms using an existing visualization software prototype. The expert indicated that a refined version of coevolutionary GA known as PAMDGA was found to most reliably locate subnetworks containing catch differences of biological relevance.