@InProceedings{ ho.ea:explainable:2026,
  title    = {Explainable Security Investment: A Shapley Value Inspired
              Metric},
  author   = {S. Destyny Ho and Yunxiao Zhang and Achim D. Brucker},
  note     = {14th EAI International Conference on Game Theory for
              Networks, GameNets 2025 ; Conference date: 17-03-2025 Through
              18-03-2025},
  year     = {2026},
  month    = {mar},
  areas    = {security},
  abstract = {Many existing methods can effectively find the optimal
              cybersecurity investment, but communicating these findings to
              non-technical stakeholders is a well-known cybersecurity
              challenge [27]. This work aims to provide additional metrics
              that grant further insight and justifications for an
              implemented cybersecurity portfolio. The Shapley value is a
              classic concept in cooperative game theory that quantifies the
              fair contribution of each player to a collective outcome. In
              security games, it offers a natural way to measure the
              contribution of individual security controls to overall
              defence. However, the Harsanyi dividend of combining two
              coalitions when taking their respective security reductions as
              their portfolio contribution tends to negative attributions in
              undesirable situations. This undermines interpretability. In
              this work, we propose a novel measure tailored for security
              games which guarantees non-negative Harsanyi dividends when
              combining coalitions that result in the significant increase
              of security. The method is grounded in the composition of two
              key factors that capture each controls marginal impact. These
              factors offer an interpretable and fair decomposition of the
              overall security effectiveness. Beyond interpretability, we
              demonstrate how these attributions can support decision-making
              in cyber defence planning. To address computational
              scalability, we present an approximation algorithm that
              significantly reduces runtimes with little impact on
              explainability.},
}