Federated Security Framework: A Heterogeneous Federated Learning Architecture for Privacy‐Preserving Intrusion Detection in IoT Networks

ABSTRACT Internet of things (IoT) applications are growing at an accelerating pace in the various infrastructures of critical systems, with cyber threats growing as well, making intrusion detection systems (IDS) significant. Some reflections of the traditional centralized IDS systems include privacy of data, lag time and low scalability of distributed networks. In this paper, a new framework federated security framework (FSF), a fresh model of federated learning, is introduced, which is an intrusion detector that integrates local ensemble learning and global feature extraction to improve the security of the network. The FSF uses the models of the random forest and XGBoost to issue local decisions concerning the IoT devices, whereas global 1D‐CNN carries out the extracting of the features centrally without jeopardizing the privacy of the information. Much of the literature on studying three benchmark datasets underpins the fact that XGBoost has an immensely high performance with an accuracy of 98.09% on NSL‐KDD, 97.94% on TON‐IoT, and 96.15% on CICEVSE2024. The architecture saved the communication overhead that was 67.3% of centralized methods and enhanced the computing performance by 89.2%. The performance of model distillation methods proved the accuracy of 95.65% and complexity reduction of 78%; this means that the paradigm has potential applications in supporting resource‐limited IoT applications. It is a federated solution which offers privative‐aware intrusion diagnosis solutions which are required in the contemporary distributed network infrastructure.