FedCognis: An Adaptive Federated Learning Framework for Secure Anomaly Detection in Industrial IoT-Enabled Cognitive Cities

Abstract

FedCognis is a secure and scalable federated learning framework designed for continuous anomaly detection in Industrial Internet of Things-enabled Cognitive Cities (IIoTCC). It introduces two key innovations: a Quantum Secure Authentication (QSA) mechanism for adversarial defense and integrity validation, and a Self-Attention Long Short-Term Memory (SALSTM) model for high-accuracy spatiotemporal anomaly detection. Addressing core challenges in traditional Federated Learning (FL)—such as model poisoning, communication overhead, and concept drift—FedCognis integrates dynamic trust-based aggregation and lightweight cryptographic verification to ensure secure, real-time operation across heterogeneous IIoT domains including utilities, public safety, and traffic systems. Evaluated on the WUSTL-IIoTCC-2021 dataset, FedCognis achieves 94.5% accuracy, 0.941 AUC for precision-recall, and 0.896 ROC-AUC, while reducing bandwidth consumption by 72%. The framework demonstrates sublinear computational complexity and a resilience score of 96.56% across six security dimensions. These results confirm FedCognis as a robust and adaptive anomaly detection solution suitable for deployment in large-scale cognitive urban infrastructures.