Solving the unsupervised graph partitioning problem with genetic algorithms: Classical and new encoding representations

Abstract

The Graph Partitioning Problem (GPP) is one of the most ubiquitous models that operations research practitioners encounter. Therefore, several methods have been proposed to solve it. Among these methods, Genetic Algorithm (GA) appears to carry very promising performances. However, despite the huge number of papers being published with this approach, only few of them deal with the encoding representation and its role in the reported performances. In this paper, we present classical and new encoding representations for the unsupervised graph partitioning problem. That is, we suppose that the number of partition subsets (clusters) is not known apriori. Next, we conduct an empiric comparison to identify the most promising encodings.