Differentially-private federated intrusion detection via knowledge distillation in third-party IoT systems of smart airports

Abstract

With the increasing deployment of IoT and Industry 4.0, the federated learning system was presented to preserve the privacy between the third-party IoT systems and the security operation center in smart airports. Nonetheless, the extremely skewed distribution of cyber threats increases the complexity of intrusion detection system (IDS) in smart airports, while privacy preservation limits the utility of IDS in the process of server model update. In this article, we have devised a knowledge distillation (KD)-based Convolutional Neural Network and Gated Recurrent Unit (CNN-GRU) model to improve the accuracy of multiple intrusion detection. In addition, the tradeoff between privacy and accuracy is achieved by denoising the adaptive parameter update mechanism to upgrade the optimizer of Differentially-Private (DP) Federated IDS. The results indicate high effectiveness and robustness of DP Federated KD-based IDS for third-party IoT systems of a smart airport.