ICDIS 2018

Approximating the Shapley value for the analysis of large terror networks

Roy Lindelauf

Netherlands Defence Academy, Netherlands

Abstract: We present an analysis of the extended WTC 9/11 network of Krebs  (Connections 24(3):43–52, 2002) that consists of 69 members by use of game theoretical power indices. In particular we focus on the Shapley value. It is one of the most prominent one-point solution concepts in cooperative game theory that divides revenues (or cost, power) that can be obtained by cooperation of players in the game. With respect to terrorism the revenues can be seen to coincide with contextual models of terror related phenomena. Since the computational complexity increases exponentially with the number of players in the game, computing the Shapley value for games consisting of more than 25 players becomes too time expensive. Therefore a new method to approximate the Shapley value by refining the random sampling method introduced by Castro et al.. We show that our method outperforms the random sampling method, reducing the average error in the Shapley value approximation by almost 30%.