Cyber Attack Detection Using Deep Multi-agent Reinforcement Learning with Beth Dataset

The Internet of Things (IoT) transforms connectivity, enabling devices to share data seamlessly via the cloud or wireless networks. However, IoT systems remain vulnerable to sophisticated cyberattacks, and traditional detection methods often fail due to inadequate fine-tuning and ineffective feature extraction. This study addresses these challenges by proposing a novel cyberattack detection framework based on deep multi-agent reinforcement learning (MARL). The framework employs the deep deterministic policy gradient (DDPG) algorithm to enhance the adaptive capabilities of multiple agents within an IoT network environment. A newly curated dataset, BETH, is introduced, encompassing diverse cyberattack scenarios specifically designed for training and evaluating MARL-based systems. Through deep reinforcement learning, the framework autonomously learns to detect and mitigate known and emerging threats. To assess the robustness and generalizability of the proposed model, a fivefold cross-validation process was conducted. Experimental results reveal that the proposed approach achieves an exceptional detection accuracy of 99.95%, demonstrating its ability to adapt to dynamic attack patterns while minimizing error rates. The findings highlight the framework’s potential to improve network security significantly, advancing the field of intelligent, autonomous cyber defence for IoT systems.