An Adaptive CGAN-Enhanced Federated Learning Model for Intrusion Detection in IoT
Hao Shen, Jingwei Hai, Jiaran Guo, Yang Cao, Yongli Yang, Shiqiang Zhang · 2024
In traditional IoT intrusion detection systems, data is typically stored and processed centrally, which can lead to privacy breaches and data security issues. In contrast, in conventional distributed IoT intrusion detection systems, the local traffic data and label distribution across different IoT devices are often imbalanced, which can result in suboptimal model training and, consequently, lower detection accuracy. To address the issue, our paper proposes an Adaptive Federated Intrusion Detection System based on Generative Adversarial Networks (AFCI). By combining Federated Learning (FL) with Intrusion Detection Systems (IDS), the proposed approach effectively solves the privacy concerns associated with device data. Additionally, Generative Adversarial Networks (GANs) excel at data augmentation, which helps mitigate the imbalance in local data across IoT devices. Extensive experiments conducted on NSL-KDD assess the AFCI model, and the results demonstrate that it outperforms existing federated intrusion detection systems in terms of performance.