Finding balanced graph bi-partitions using a hybrid genetic algorithm

Abstract

Proposes a hybrid genetic algorithm (GA) for the graph-balanced bi-partition problem, a challenging NP-hard combinatorial optimization problem arising in many practical applications. The hybrid character of the GA lies in the application of a heuristic procedure to improve candidate solutions. The basic idea behind our heuristic is to identify and exploit clusters, i.e. subgraphs with a relatively high edge density. The resulting hybrid genetic algorithm turns out to be very effective, both in terms of quality of solutions and running time. On a large class of benchmark families of graphs, our hybrid genetic algorithm yields results of the same or better quality than those obtained by all other heuristic algorithms we are aware of, for comparable running times.