Scholarlink

<p>The rapid advancements in Internet technologies and communication systems have led to a dramatic increase in network size and data volume. This rapid growth has simultaneously introduced a wide range of new and sophisticated cyber threats. Among these threats, distributed denial-of-service attacks are particularly destructive, posing severe challenges to the availability and integrity of network infrastructures. Although deep learning (DL) models have achieved remarkable success in intrusion detection, their deterministic nature limits their performance under uncertain or noisy network conditions. To overcome this limitation, this paper proposes a hybrid intrusion detection framework that integrates convolutional neural network (CNN) architectures—VGG16 and VGG19—with a fuzzy inference mechanism to enhance decision reliability. In the proposed method, tabular network traffic records from the CICDDoS2019 and UNSW_NB15 datasets are first converted into image-like representations, enabling the use of pretrained VGG models for hierarchical feature extraction. The proposed framework fuses the probabilistic outputs of both VGG networks using a fuzzy logic layer to explicitly model uncertainty and refine the classification boundaries. In addition, the proposed framework has been developed with compatibility for high-performance computing (HPC) environments. The parallel nature of convolutional neural networks, combined with the independent inference of VGG16 and VGG19 models, enables efficient execution on GPU-based and distributed computing platforms. These characteristics make the proposed approach suitable for large-scale and real-time DDoS detection scenarios, where high throughput and low latency are essential requirements. Binary classification and multiclassification are performed to evaluate the model comprehensively. Experimental evaluation on the CICDDoS2019 dataset demonstrates that the hybrid VGG–fuzzy model achieves 99.79% accuracy in binary classification and 98.76% in multiclassification, outperforming individual CNN and fuzzy systems as well as comparable state-of-the-art methods. The results indicate that combining deep feature extraction with fuzzy reasoning significantly improves detection robustness, offering a scalable and adaptive approach for real-world DDoS mitigation.</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 decision-level hybrid VGG–fuzzy framework for robust and interpretable DDoS attack detection in network traffic

  • 期刊论文

  • 发表日期：

    2026年4月20日

查看全文
  • Roya Jamaati Gashti,
  • Mehdi Sadeghzadeh,
  • Alireza TamjidYamcholo,
  • Ladan Riazi

摘要

The rapid advancements in Internet technologies and communication systems have led to a dramatic increase in network size and data volume. This rapid growth has simultaneously introduced a wide range of new and sophisticated cyber threats. Among these threats, distributed denial-of-service attacks are particularly destructive, posing severe challenges to the availability and integrity of network infrastructures. Although deep learning (DL) models have achieved remarkable success in intrusion detection, their deterministic nature limits their performance under uncertain or noisy network conditions. To overcome this limitation, this paper proposes a hybrid intrusion detection framework that integrates convolutional neural network (CNN) architectures—VGG16 and VGG19—with a fuzzy inference mechanism to enhance decision reliability. In the proposed method, tabular network traffic records from the CICDDoS2019 and UNSW_NB15 datasets are first converted into image-like representations, enabling the use of pretrained VGG models for hierarchical feature extraction. The proposed framework fuses the probabilistic outputs of both VGG networks using a fuzzy logic layer to explicitly model uncertainty and refine the classification boundaries. In addition, the proposed framework has been developed with compatibility for high-performance computing (HPC) environments. The parallel nature of convolutional neural networks, combined with the independent inference of VGG16 and VGG19 models, enables efficient execution on GPU-based and distributed computing platforms. These characteristics make the proposed approach suitable for large-scale and real-time DDoS detection scenarios, where high throughput and low latency are essential requirements. Binary classification and multiclassification are performed to evaluate the model comprehensively. Experimental evaluation on the CICDDoS2019 dataset demonstrates that the hybrid VGG–fuzzy model achieves 99.79% accuracy in binary classification and 98.76% in multiclassification, outperforming individual CNN and fuzzy systems as well as comparable state-of-the-art methods. The results indicate that combining deep feature extraction with fuzzy reasoning significantly improves detection robustness, offering a scalable and adaptive approach for real-world DDoS mitigation.