Scholarlink

<p>The rapidly expanding Agricultural Internet of Things (Ag-IoT) sector enhances efficiency and reliability and increases cyber vulnerabilities. Although deep learning and machine learning-based intrusion detection systems (IDS) address security issues, they struggle with centralized data collection and privacy concerns. Federated learning offers a solution that achieves high detection rates without centralizing data. However, real-world implementation faces challenges such as limited IoT device resources and communication overhead associated with parameter transmission. Lightweight models with fewer parameters and lower memory and energy consumption are crucial for Ag-IoT devices, which require shorter training and detection times. This paper presents a lightweight federated learning-based IDS that combines Random Forest for feature selection and Temporal Convolutional Network (TCN) for deep learning classification. Tested on the UNSW-NB15, Edge-IIoT, and CIC-IoT23 data sets with FedAvg, FedProx, and FedAVGM aggregators, the proposed IDS is more efficient in training and detection, lighter than commonly used models, and achieves superior attack-detection performance.</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

Privacy-preserving lightweight intrusion detection system for agricultural IoT based on federated learning and temporal convolutional network

  • 期刊论文

  • 发表日期：

    2026年7月31日

查看全文
  • Amina Khacha,
  • Zibouda Aliouat,
  • Yasmine Harbi,
  • Chirihane Gherbi,
  • Rafika Saadouni,
  • Hicham Lakhlef

摘要

The rapidly expanding Agricultural Internet of Things (Ag-IoT) sector enhances efficiency and reliability and increases cyber vulnerabilities. Although deep learning and machine learning-based intrusion detection systems (IDS) address security issues, they struggle with centralized data collection and privacy concerns. Federated learning offers a solution that achieves high detection rates without centralizing data. However, real-world implementation faces challenges such as limited IoT device resources and communication overhead associated with parameter transmission. Lightweight models with fewer parameters and lower memory and energy consumption are crucial for Ag-IoT devices, which require shorter training and detection times. This paper presents a lightweight federated learning-based IDS that combines Random Forest for feature selection and Temporal Convolutional Network (TCN) for deep learning classification. Tested on the UNSW-NB15, Edge-IIoT, and CIC-IoT23 data sets with FedAvg, FedProx, and FedAVGM aggregators, the proposed IDS is more efficient in training and detection, lighter than commonly used models, and achieves superior attack-detection performance.