Scholarlink

An Intrusion Detection System (IDS) plays an important role in securing the network from harmful cyber-attacks. In this paper, a novel framework is introduced that combines federated ensemble learning, unsupervised clustering, and advanced feature selection to develop a lightweight and privacy-preserving IDS. The proposed approach is evaluated on benchmark datasets such as NSL-KDD and UNSW-NB15. A Modified Shuffled Frog Leaping Algorithm (SFLA) is introduced for optimal feature selection, utilizing ensemble classifiers (MLP, Random Forest, and Xgboost) and a new fitness function that balances the False Alarm Rate (FAR) and F1-score. After feature selection, Fuzzy C-Means (FCM) clustering is applied to attack samples to form multiple clients for Horizontal Federated Learning (HFL). Each client trains locally while sharing only feature importance values for global aggregation, ensuring data privacy. Finally, a soft voting–based ensemble model is used for prediction. Accuracy rate of proposed model is 99.28% and 91.04% on the NSL-KDD and UNSW-NB15 datasets, respectively. Experimental results confirm that the proposed system effectively detects diverse attacks in a decentralized environment while maintaining high accuracy, low false alarm rate, and strong data privacy.

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

Designing Lightweight and Privacy-Preserving IDS Applying Federated Learning Technique

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Aditya Kumar Singh,
  • Arpita Srivastava,
  • Ditipriya Sinha

摘要

An Intrusion Detection System (IDS) plays an important role in securing the network from harmful cyber-attacks. In this paper, a novel framework is introduced that combines federated ensemble learning, unsupervised clustering, and advanced feature selection to develop a lightweight and privacy-preserving IDS. The proposed approach is evaluated on benchmark datasets such as NSL-KDD and UNSW-NB15. A Modified Shuffled Frog Leaping Algorithm (SFLA) is introduced for optimal feature selection, utilizing ensemble classifiers (MLP, Random Forest, and Xgboost) and a new fitness function that balances the False Alarm Rate (FAR) and F1-score. After feature selection, Fuzzy C-Means (FCM) clustering is applied to attack samples to form multiple clients for Horizontal Federated Learning (HFL). Each client trains locally while sharing only feature importance values for global aggregation, ensuring data privacy. Finally, a soft voting–based ensemble model is used for prediction. Accuracy rate of proposed model is 99.28% and 91.04% on the NSL-KDD and UNSW-NB15 datasets, respectively. Experimental results confirm that the proposed system effectively detects diverse attacks in a decentralized environment while maintaining high accuracy, low false alarm rate, and strong data privacy.