Robust Defense Against Data Reconstruction Attack in Federated Industrial Intrusion Detection Systems
Areeb Ahmed Bhutta, Adnan Noor Mian · IEEE Transactions on Industrial Informatics · 2025
Industrial intrusion detection systems (IIDS) are crucial in defending digital industrial infrastructures by detecting unauthorized activities within industrial network traffic. As cyber threats continue to evolve, collaborative advancements in IIDS are necessary to ensure robust defenses across future industries. While machine learning (ML) enhances IIDS capabilities, collaborative ML approaches face privacy and regulatory challenges, limiting data sharing across industries. Federated learning (FL) offers a solution by enabling collaborative training without direct data sharing; however, industries remain vulnerable to data reconstruction attacks and FL can impose high overheads. To address these challenges, we propose federated neural-network-based gradient boosting (FNGB) method for collaborative IIDS. FNGB introduces GradProtect, a privacy-preserving mechanism that mitigates data reconstruction, and DynamicLR, an adaptive learning rate method for efficient distributed gradient boosting. Extensive evaluations demonstrate that FNGB delivers enhanced privacy, high performance, and high convergence with low overhead compared to existing methods, making it well-suited for cross-industry deployment.