Scholarlink

<p>Distributed denial of service attacks are increasingly targeting Internet of Things devices, threatening user privacy, data security, and system stability. Despite efforts to enhance DDoS detection in IoT environments, existing methods struggle with accuracy and real-time performance when processing large-scale network traffic. This study introduces time-enhanced transformer for security, an enhanced ContiFormer model integrated with feature selection for network intrusion detection. The ContiFormer employs a continuous-time Transformer architecture, excelling at processing irregularly timed IoT network traffic. TETS optimizes temporal encoding to accurately detect anomalous patterns in data streams. Additionally, the pigeonhole optimization algorithm, selected from six feature selection methods for its superior performance, extracts key features from massive network data, reducing computational complexity and noise while boosting detection accuracy and efficiency. TETS achieves detection accuracies of 99.45%, 99.40%, 99.46%, and 96.26% on the CICDDoS2019, CICIDS2018, UNSW-NB15-v2, and IoT2023 datasets, respectively, outperforming baseline CNN, Transformer, and other models.</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

A DDoS attack detection method based on improved transformer and temporal feature enhancement

  • 期刊论文

  • 发表日期：

    2025年6月2日

查看全文
  • Yifan Fan,
  • Hao Ma,
  • Yiying Zhang,
  • Siwei LI,
  • Xiaoyan Guo,
  • Ben Wang

摘要

Distributed denial of service attacks are increasingly targeting Internet of Things devices, threatening user privacy, data security, and system stability. Despite efforts to enhance DDoS detection in IoT environments, existing methods struggle with accuracy and real-time performance when processing large-scale network traffic. This study introduces time-enhanced transformer for security, an enhanced ContiFormer model integrated with feature selection for network intrusion detection. The ContiFormer employs a continuous-time Transformer architecture, excelling at processing irregularly timed IoT network traffic. TETS optimizes temporal encoding to accurately detect anomalous patterns in data streams. Additionally, the pigeonhole optimization algorithm, selected from six feature selection methods for its superior performance, extracts key features from massive network data, reducing computational complexity and noise while boosting detection accuracy and efficiency. TETS achieves detection accuracies of 99.45%, 99.40%, 99.46%, and 96.26% on the CICDDoS2019, CICIDS2018, UNSW-NB15-v2, and IoT2023 datasets, respectively, outperforming baseline CNN, Transformer, and other models.