Scholarlink

<p>Semi-flooding (SF) and low-rate (LR) Distributed Denial of Service (DDoS) attacks can be particularly hard to counter and sustain longer than high-rate attacks. The main reason is that they resemble regular network traffic, making it difficult to spot and respond to them. This study introduces an innovative approach to handle DDoS attacks in Internet of Things (IoT) setups that use Software-Defined Networking (SDN). The proposed method uses an entropy-based deep learning (EnDL) model with an adaptable network structure to detect, decide, and defend against SF and LR attacks effectively. EnDL draws on approximate entropy (ApEn) and sample entropy (SampEn) to study traffic patterns and employs DL to refine error detection. The core of the DL model integrates convolutional neural networks (CNN), attention mechanisms, and a Bidirectional Gated Recurrent Unit (BiGRU). EnDL employs adaptive topology by reshaping the structure and device entropy to regularly update the network’s routing paths to disrupt attack attempts. It also organizes IoT devices into specific clusters based on location, identity, and type. This setup allows network managers to focus precisely on critical areas for traffic analysis. Tests conducted on two standard datasets (CIC-DDoS2023 and Edge-IoT) and a custom attack scenario showed that EnDL provides improved detection accuracy, a lower false positive rate (FPR), and quicker recovery times after an attack. Compared to other solutions, EnDL performs 0.3–13% better in these areas.</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

Approximate Entropy-Based DL Model with Adaptive Topology and Clustering Restriction for Semi-Flooding and Low-rate Attack Detection and Mitigation

  • 期刊论文

  • 发表日期：

    2026年4月13日

查看全文
  • Mohammad Hormozi,
  • Seyed Hossein Erfani,
  • Amir Sahafi,
  • Mona Moradi

摘要

Semi-flooding (SF) and low-rate (LR) Distributed Denial of Service (DDoS) attacks can be particularly hard to counter and sustain longer than high-rate attacks. The main reason is that they resemble regular network traffic, making it difficult to spot and respond to them. This study introduces an innovative approach to handle DDoS attacks in Internet of Things (IoT) setups that use Software-Defined Networking (SDN). The proposed method uses an entropy-based deep learning (EnDL) model with an adaptable network structure to detect, decide, and defend against SF and LR attacks effectively. EnDL draws on approximate entropy (ApEn) and sample entropy (SampEn) to study traffic patterns and employs DL to refine error detection. The core of the DL model integrates convolutional neural networks (CNN), attention mechanisms, and a Bidirectional Gated Recurrent Unit (BiGRU). EnDL employs adaptive topology by reshaping the structure and device entropy to regularly update the network’s routing paths to disrupt attack attempts. It also organizes IoT devices into specific clusters based on location, identity, and type. This setup allows network managers to focus precisely on critical areas for traffic analysis. Tests conducted on two standard datasets (CIC-DDoS2023 and Edge-IoT) and a custom attack scenario showed that EnDL provides improved detection accuracy, a lower false positive rate (FPR), and quicker recovery times after an attack. Compared to other solutions, EnDL performs 0.3–13% better in these areas.