Scholarlink

<p>Integrating Flying Ad hoc Network (FANET) technology with smart sensors and devices has sparked numerous innovative applications. However, this integration has exposed FANETs to significant security threats, undermining network integrity and reliability. Deep learning (DL) has recently garnered attention for intrusion detection systems (IDSs) within FANETs. While centralized DL (CDL) offers robustness and effectiveness, it raises concerns about privacy, computational overhead, and scalability. In contrast, federated deep learning (FDL) ensures data privacy and conserves drone energy, but faces deployment challenges in dynamic and resource-constrained FANETs, including communication constraints and drone heterogeneity. To address these challenges, optimal drone clients must be selected for participation in FDL, necessitating precise drone positioning. This paper proposes a reinforcement learning-based drone client selection algorithm to determine optimal drone-client participation in FDL training. Building upon this algorithm, we introduce a novel FDL-based IDS (FLID) for efficient detection of malicious attacks in FANETs with minimal computational overhead and data-privacy preservation, addressing communication and drone heterogeneity constraints. FLID incorporates multilayer perceptron, convolutional neural networks, and recurrent neural networks with long short-term memory using the WSN-BFSF dataset tailored for detecting flooding, blackhole, and selective forwarding attacks. Experimental results demonstrate that FLID fulfills IDS requirements and achieves superior performance compared to other reference models, with an accuracy of 99.38%, precision of 99%, recall of 99%, and an F1-score of 99%.</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

Reinforcement learning-based drone-client selection for efficient federated learning-based intrusion detection in FANETs

  • 期刊论文

  • 发表日期：

    2025年8月23日

查看全文
  • Sihem Benfriha,
  • Nabila Labraoui,
  • Haythem A. Bany Salameh,
  • Radjaa Bensaid

摘要

Integrating Flying Ad hoc Network (FANET) technology with smart sensors and devices has sparked numerous innovative applications. However, this integration has exposed FANETs to significant security threats, undermining network integrity and reliability. Deep learning (DL) has recently garnered attention for intrusion detection systems (IDSs) within FANETs. While centralized DL (CDL) offers robustness and effectiveness, it raises concerns about privacy, computational overhead, and scalability. In contrast, federated deep learning (FDL) ensures data privacy and conserves drone energy, but faces deployment challenges in dynamic and resource-constrained FANETs, including communication constraints and drone heterogeneity. To address these challenges, optimal drone clients must be selected for participation in FDL, necessitating precise drone positioning. This paper proposes a reinforcement learning-based drone client selection algorithm to determine optimal drone-client participation in FDL training. Building upon this algorithm, we introduce a novel FDL-based IDS (FLID) for efficient detection of malicious attacks in FANETs with minimal computational overhead and data-privacy preservation, addressing communication and drone heterogeneity constraints. FLID incorporates multilayer perceptron, convolutional neural networks, and recurrent neural networks with long short-term memory using the WSN-BFSF dataset tailored for detecting flooding, blackhole, and selective forwarding attacks. Experimental results demonstrate that FLID fulfills IDS requirements and achieves superior performance compared to other reference models, with an accuracy of 99.38%, precision of 99%, recall of 99%, and an F1-score of 99%.