Toward more effective centrality-based attacks on network topologies

This paper considers the cyber-attacks that aim to remove nodes or links from network topologies. We particularly focus on one category of such attacks, in which attacks happen by rounds, and in each round, the node with the highest centrality and its adjacent links are removed. Here the centrality can be any centrality measure such as Degree Centrality, Betweenness Centrality, etc. For this attack category, there currently exist two strategies: Initial and Adaptive. In the Initial strategy, node centralities are only calculated initially, while in the Adaptive strategy, node centralities are recalculated after each round of attack. In the literature, it has been shown that the Adaptive strategy is more effective than the Initial strategy for a centrality measure. In this paper, we propose a new strategy called the largest component (LC) strategy which further outperforms the Adaptive strategy in terms of both attack effectiveness and computation complexity. Moreover, we propose the use of current-flow versions of Betweenness Centrality and Closeness Centrality as the centrality measures in the attacks, since they are more granular and supported by the LC strategy. We verify the better performances of the LC strategy by extensive experiments on four kinds of artificial networks and two realworld networks. Our experiments also show that the Currentflow Betweenness Centrality makes attacks the most effective among the five centrality measures studied in this paper