Federated learning with LSTM and error correcting codes for secure and private identification of IoT devices

The rapid proliferation of Internet-of-Things (IoT) devices has enabled seamless automation and data-driven intelligence across diverse domains, including smart homes, healthcare, and industrial systems. However, the resulting hyper-connectivity also expands the attack surface, raising critical concerns regarding privacy, device identification, and network security. This paper introduces FL-HDECOC, a novel federated learning-based framework for privacy-preserving device identification in heterogeneous IoT environments. The proposed model combines the temporal modeling capabilities of Long Short-Term Memory (LSTM) networks with the robustness of Error-Correcting Output Codes (DECOC) for multi-class classification, integrated within a federated architecture that ensures data locality and user privacy. Differential privacy is further employed to formally protect shared model updates against inference attacks. Experimental evaluations demonstrate that FL-HDECOC outperforms baseline models, achieving 96.2% accuracy and 95.5% F1-score, while reducing communication rounds by 25% under privacy budget ε = 1.0. These results highlight its promise for secure and scalable IoT deployments.