Scholarlink

<p>Intrusion detection in heterogeneous edge and IoT environments is challenging due to resource-constrained devices, non-IID data distributions, and the need to preserve data privacy without centralized data sharing. To address these challenges, this paper presents FL–ILN, a federated intrusion detection framework that integrates dynamic programming-based client selection, an Improved LinkNet model, median aggregation, and a hybrid feature representation using statistical and raw features. The framework also employs a modified quantile loss to improve convergence stability and handle class imbalance during distributed training. Experiments on the UNSW-NB15 dataset demonstrate that the proposed framework achieves strong intrusion detection performance, with 0.968 accuracy, 0.028 false positive rate, and a 5.9% mean accuracy improvement over DenseNet. These results indicate that FL–ILN provides an effective, robust, and communication-efficient solution for privacy-preserving intrusion detection in edge computing environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Federated learning-enabled intrusion detection framework utilizing dynamic programming based client selection and an improved linknet model in edge computing environment

  • 期刊论文

  • 发表日期：

    2026年6月29日

查看全文
  • Vijay Singh,
  • Bharadwaj Veeravalli

摘要

Intrusion detection in heterogeneous edge and IoT environments is challenging due to resource-constrained devices, non-IID data distributions, and the need to preserve data privacy without centralized data sharing. To address these challenges, this paper presents FL–ILN, a federated intrusion detection framework that integrates dynamic programming-based client selection, an Improved LinkNet model, median aggregation, and a hybrid feature representation using statistical and raw features. The framework also employs a modified quantile loss to improve convergence stability and handle class imbalance during distributed training. Experiments on the UNSW-NB15 dataset demonstrate that the proposed framework achieves strong intrusion detection performance, with 0.968 accuracy, 0.028 false positive rate, and a 5.9% mean accuracy improvement over DenseNet. These results indicate that FL–ILN provides an effective, robust, and communication-efficient solution for privacy-preserving intrusion detection in edge computing environments.