Efficiency of attack strategies on complex model and real-world networks

Abstract

We investigated the efficiency of attack strategies to network nodes when targeting several complex model and real-world networks. We tested 5 attack strategies, 3 of which were introduced in this work for the first time, to attack 3 model (Erdos and Renyi, Barabasi and Albert preferential attachment network, and scale-free network configuration models) and 3 real networks (Gnutella peer-to-peer network, email network of the University of Rovira i Virgili, and immunoglobulin interaction network). Nodes were removed sequentially according to the importance criterion defined by the attack strategy. We used the size of the largest connected component (LCC) as a measure of network damage. We found that the efficiency of attack strategies (fraction of nodes to be deleted for a given reduction of LCC size) depends on the topology of the network, although attacks based on the number of connections of a node and betweenness centrality were often the most efficient strategies. Sequential deletion of nodes in decreasing order of betweenness centrality was the most efficient attack strategy when targeting real-world networks. In particular for networks with power-law degree distribution, we observed that most efficient strategy change during the sequential removal of nodes.