Quantum-safe Federated Learning Framework for Cyber Threat Detection in 5G-enabled IoT Ecosystems

Internet of Things (IoT) ecosystems refer to interconnected intelligent devices that enhance automation and efficiency by enabling seamless data exchange. IoT ecosystem data, generated through intelligent networks like 5G, is growing exponentially. However, its data island nature hinders secure aggregation. Furthermore, cloud-trained models struggle to provide personalized security solutions, which amplifies security and privacy challenges due to increased vulnerability to cyber threats. To address these challenges, deep Federated Learning (FL) solves data island issues without sharing private data, while Cyber Threat Intelligence (CTI) employs AI models to understand and detect cyber threats to protect 5G-enabled IoT ecosystems, which addresses personalization concerns. However, much literature indicates that the privacy of classical and quantum systems can still be compromised during the communication process in deep FL. In addition, the training cost of the local CTI model is also relatively high in deep FL. This study first developed a novel lightweight deep learning-based CTI model that integrates MobileNet-V1 with a bidirectional gated recurrent unit to detect cyber threats. Second, an FL framework is designed to enable multiple IoT clients with heterogeneous data to collectively build a comprehensive CTI model, while preserving privacy and personalized FL. Third, a post-quantum cryptographic protocol based on Learning with Errors is crafted to secure model parameters during client-server communication. Extensive experiments on real-world 5G-NIDD datasets demonstrate the exceptional performance of the proposed model in detecting multiple cyber threats targeting IoT ecosystems, achieving 99.82% accuracy and surpassing contemporary methods. This framework substantially improves protection for the vital 5G-enabled IoT ecosystem against emerging cyber threats in the quantum era.