Secure Federated Maximum Entropy Classifier for Intrusion Detection in IoT Consumer Electronics

The rapid proliferation of connected consumer electronics within the Internet of Things (IoT) ecosystem has enabled a wide range of smart applications, but it has also exposed devices and networks to significant cybersecurity threats. Conventional intrusion detection systems based on centralized data collection suffer from critical limitations in terms of privacy preservation, scalability, and latency when deployed in large-scale IoT environments. To address these challenges, this paper proposes a Secure Federated Maximum Entropy Classifier (SFMEC) for IoT intrusion detection. The proposed SFMEC framework leverages federated learning to collaboratively train intrusion detection models across IoT gateways and edge servers, thereby enhancing scalability while preserving data privacy. Raw data generated by registered IoT devices is encrypted using agreed session keys and transmitted to their associated gateways, where local models are trained using a maximum entropy classifier. To improve security and robustness, both local and global models are protected using symmetric-key encryption and compressed using the ZIP algorithm to prevent parameter leakage and mitigate adversarial manipulation during model communication between IoT gateways and edge servers. Extensive experiments conducted on benchmark IoT intrusion detection datasets demonstrate that SFMEC consistently achieves higher detection accuracy and lower false-positive rates compared to existing approaches. The results highlight that combining entropy-based learning with secure federated optimization provides resilience against evolving threats and communication overheads; therefore, SFMEC enhances the security posture of next-generation IoT ecosystems.