Adaptive Honeypot Deployment Strategy for IoT Networks Using Reinforcement Learning and Adversarial Game Theory

The proliferation of Internet of Things (IoT) devices across critical infrastructure and smart environments has introduced significant security vulnerabilities, primarily due to device heterogeneity, limited computational resources, and pervasive weak configurations. While honeypots serve as effective deception mechanisms to detect intrusions and gather threat intelligence, static honeypot deployments are easily identified and bypassed by adaptive adversaries. In this paper, we propose a novel adaptive honeypot deployment framework for IoT networks that integrates Bayesian Stackelberg game theory with deep reinforcement learning (DRL). We model the strategic interaction between a network defender and an adaptive attacker as a dynamic game with incomplete information, where the defender dynamically optimizes the placement, interaction level, and resource allocation of virtualized IoT honeypots. A Double Deep Q-Network (DDQN) agent is implemented to solve the game dynamically, learning optimal deception policies in response to evolving attacker reconnaissance behaviors. Evaluated in a high-fidelity simulated IoT environment under diverse multi-stage attack scenarios, our framework demonstrates a 34.6% increase in attacker interception rates and a 42.1% reduction in unauthorized network discovery compared to baseline static and heuristic deployment strategies, while maintaining negligible network overhead.