Cutting Graphs Using Competing Ant Colonies and an Edge Clustering Heuristic

We investigate the usage of Ant Colony Optimization to detect balanced graph cuts. In order to do so we develop an algorithm based on competing ant colonies. We use a heuristic from social network analysis called the edge clustering coefficient, which greatly helps our colonies in local search. The algorithm is able to detect cuts that correspond very well to known cuts on small real-world networks. Also, with the correct parameter balance, our algorithm often outperforms the traditional Kernighan-Lin algorithm for graph partitioning with equal running time complexity. On larger networks, our algorithm is able to obtain low cut sizes, but at the cost of a balanced partition.