ePrints

EBMADDPG: Shapley-based explainable moving target defense for edge intelligence-enabled SIoT systems via joint Bayesian Markov games and DRL

Lookup NU author(s): Yizhou Shen

Downloads

Full text for this publication is not currently held within this repository. Alternative links are provided below where available.

Abstract

© 2025 Elsevier B.V.The edge intelligence (EI)-enabled social Internet of Things (SIoT) is increasingly vulnerable to sophisticated malware that exploits social relationships between devices to propagate rapidly and bypass traditional security. To address this dynamic threat under incomplete information, we propose a novel moving target defense framework based on a Bayesian Markov game. In our framework, defenders dynamically shift system configurations and resource allocations upon detecting potential threats. Based on their belief states about attacker types, each defender can decide whether to coordinate defense strategies with other agents. Unlike most existing work, we explicitly account for both the incomplete information about attacker capabilities and the dynamic nature of EI-enabled SIoT systems. We formulate a joint optimization problem to simultaneously determine belief updates about attacker types via Bayesian inference, dynamic reconfiguration of defense parameters, and optimal coordination strategies among agents. To efficiently solve this problem, we develop a novel explainable bayesian multi-agent deep deterministic policy gradient algorithm, which integrates centralized training with decentralized execution. Furthermore, we incorporate Shapley Additive Explanations to analyze agent contributions. Theoretical analyses and extensive simulations demonstrate that our proposed solution significantly outperforms traditional reinforcement learning algorithms.

Publication metadata

Author(s): Hong T, Shen Y, Wu X, Dong J, Shen S, Liu Z

Publication type: Article

Publication status: Published

Journal: Information Fusion

Year: 2026

Volume: 130

Print publication date: 01/06/2026

Online publication date: 27/12/2025

Acceptance date: 23/12/2025

ISSN (print): 1566-2535

ISSN (electronic): 1872-6305

Publisher: Elsevier B.V.

URL: https://doi.org/10.1016/j.inffus.2025.104101

DOI: 10.1016/j.inffus.2025.104101

Altmetrics

Share