A Federated Learning and GAN-Based Stealthy Malware Detection Scheme for IIoT Security

Industrial Internet of Things (IIoT) devices are vital for real-time data collection, yet their widespread adoption exposes networks to stealthy malware that mimics benign processes, creating major challenges for secure and efficient deployment in sensitive domains such as clinical decision-making. Effective detection is further constrained by data scarcity due to privacy concerns and restricted access to malware samples. To address these challenges, this paper proposes a Federated Stealth Malware Detection (FSMD) scheme that combines Generative Adversarial Networks (GANs) for synthetic malware augmentation with Federated Learning for privacy-preserving collaborative training. In addition, an adaptive satisfaction scoring method is introduced to reduce redundant training rounds and optimize resource usage at edge devices. Experimental results demonstrate that FSMD achieves 96.34% detection accuracy, outperforming existing methods while reducing training rounds by 36% on average compared with standard Federated Averaging (p < 0.001). Validation on Raspberry Pi hardware further confirms its efficiency under realistic IIoT constraints. By uniting data augmentation, privacy-preserving learning, and adaptive training control, FSMD provides a novel and practical solution for strengthening IIoT security against evolving stealthy malware threats.