Federated Industrial IoT Threats Detection

Abstract

Industrial Internet of Things networks are increasingly exposed to coordinated cyber threats, including denial-of-service attacks, botnet propagation, command injection, data exfiltration, false-data injection, and lateral movement across cyberphysical production systems. Conventional centralized intrusion detection models often require raw industrial traffic to be transmitted to cloud servers, creating privacy risks, bandwidth overhead, latency constraints, and weak adaptability across heterogeneous plants. This paper proposes a privacy-preserving federated threat-detection framework titled Fed-IIoTGuard, a novel hybrid algorithm that integrates temporal convolutional feature extraction, gated recurrent traffic profiling, attentionbased client weighting, and adaptive secure aggregation for detecting threats in distributed Industrial IoT environments. The proposed model is designed to learn from multiple edge gateways, programmable logic controller networks, smart sensors, supervisory control and data acquisition nodes, and industrial robots without exposing raw operational data.
Fed-IIoTGuard is compared with traditional machine learning and federated learning baselines, including Random Forest, Support Vector Machine, XGBoost, centralized CNN-LSTM, FedAvg, FedProx, FedYogi, FedNova, and SCAFFOLD. The evaluation is structured around key performance indicators such as accuracy, precision, recall, F1-score, false alarm rate, detection latency, communication overhead, convergence speed, and robustness under non-IID data distribution. The paper further presents comparative graphs showing algorithmic performance across normal traffic, denial-of-service traffic, reconnaissance attacks, command injection, spoofing, and data manipulation attacks. The proposed Fed-IIoTGuard framework is expected to demonstrate superior performance by improving minority attack detection, reducing false positives, accelerating convergence, and maintaining data confidentiality across distributed industrial sites. The study contributes a scalable and privacy-aware threat-detection architecture for next-generation smart factories, energy systems, manufacturing plants, and critical industrial infrastructures.