Scholarlink

<p>Fog computing brings cloud services closer to the network edge. This extension allows low-latency processing, making it suitable for applications such as competent healthcare and autonomous systems. However, its decentralized architecture and limited resources also introduce new security challenges. These characteristics make fog environments particularly vulnerable to advanced cyber threats, including low-rate Distributed Denial-of-Service (DDoS) attacks and cross-site scripting combined with SQL injection (XSS-SQLi). Traditional centralized Intrusion Detection Systems (IDS) often fall short in these settings. They face limitations in privacy, scalability, and adaptability, which hinder their effectiveness in fog-based scenarios. To address these challenges, this paper introduces <i>FedContinualIDS</i>—a novel intrusion detection framework based on <i>Federated Continual Learning (FCL)</i>. The goal is to enable privacy-preserving, scalable, and adaptive attack detection specifically for fog computing environments. The proposed framework integrates FCL with <i>Manta Ray Foraging Optimization (MRFO)</i> for feature selection. It uses the <i>HIKARI dataset</i>, which initially contains 555,278 samples and 85 features, including both real and synthetic attacks relevant to fog systems. Data preprocessing involves removing outliers through <i>Isolation Forest</i> and addressing class imbalance using <i>SMOTE (Synthetic Minority Oversampling Technique)</i> on local training data only. Local models are trained on individual fog nodes using <i>Elastic Weight Consolidation (EWC)</i> to mitigate catastrophic forgetting. These models are then aggregated globally using federated averaging. Experimental results show that FedContinualIDS achieves a detection accuracy of 95.67%, significantly outperforming traditional approaches. In comparison, Artificial Neural Networks (ANN) and Feedforward Neural Networks (FNN) achieve 84.74 and 87.92% accuracy, respectively. The proposed system also shows improvements in precision, recall, and F1-score, particularly for minority-class attacks. It demonstrates strong detection performance and enhanced scalability. Overall, the privacy-preserving, adaptive, and efficient design of FedContinualIDS makes it a robust security solution for fog computing. It has strong potential for real-world deployment in latency-sensitive, resource-constrained Internet of Things (IoT) 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

Intrusion detection model for secure fog computing: harnessing with federated continual learning using the HIKARI dataset to combat emerging attacks

  • 期刊论文

  • 发表日期：

    2026年3月17日

查看全文
  • Haider Kareem Mohammed,
  • B. Indrani

摘要

Fog computing brings cloud services closer to the network edge. This extension allows low-latency processing, making it suitable for applications such as competent healthcare and autonomous systems. However, its decentralized architecture and limited resources also introduce new security challenges. These characteristics make fog environments particularly vulnerable to advanced cyber threats, including low-rate Distributed Denial-of-Service (DDoS) attacks and cross-site scripting combined with SQL injection (XSS-SQLi). Traditional centralized Intrusion Detection Systems (IDS) often fall short in these settings. They face limitations in privacy, scalability, and adaptability, which hinder their effectiveness in fog-based scenarios. To address these challenges, this paper introduces FedContinualIDS—a novel intrusion detection framework based on Federated Continual Learning (FCL). The goal is to enable privacy-preserving, scalable, and adaptive attack detection specifically for fog computing environments. The proposed framework integrates FCL with Manta Ray Foraging Optimization (MRFO) for feature selection. It uses the HIKARI dataset, which initially contains 555,278 samples and 85 features, including both real and synthetic attacks relevant to fog systems. Data preprocessing involves removing outliers through Isolation Forest and addressing class imbalance using SMOTE (Synthetic Minority Oversampling Technique) on local training data only. Local models are trained on individual fog nodes using Elastic Weight Consolidation (EWC) to mitigate catastrophic forgetting. These models are then aggregated globally using federated averaging. Experimental results show that FedContinualIDS achieves a detection accuracy of 95.67%, significantly outperforming traditional approaches. In comparison, Artificial Neural Networks (ANN) and Feedforward Neural Networks (FNN) achieve 84.74 and 87.92% accuracy, respectively. The proposed system also shows improvements in precision, recall, and F1-score, particularly for minority-class attacks. It demonstrates strong detection performance and enhanced scalability. Overall, the privacy-preserving, adaptive, and efficient design of FedContinualIDS makes it a robust security solution for fog computing. It has strong potential for real-world deployment in latency-sensitive, resource-constrained Internet of Things (IoT) environments.