Scholarlink

<p>This research addresses network security by introducing a novel intrusion detection strategy, AGM, which combines the Adaptive Synthetic (ADASYN) sampling technique with a Gaussian Mixture Model (GMM) clustering approach. Traditional deep learning frameworks are enhanced by implementing the C3BANet model, which integrates Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Channel-Attention mechanisms. The study further investigates the impact of Principal Component Analysis (PCA) with varying principal component counts on model performance. Thorough preprocessing of the UNSW-NB15 and CICIDS2017 datasets was conducted to eliminate inconsistencies, and AGM sampling was applied to balance class representation. Our experiments show that the C3BANet model achieves exceptional detection rates of 99.99% and 96.82% in binary and multi-class settings on the UNSW-NB15 dataset, and 99.78% and 99.57% on the CICIDS2017 dataset, respectively. These results significantly surpass current leading algorithms, demonstrating both methodological innovation and practical advantages in intrusion detection.</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

Agm-c3banet: a network intrusion detection model for imbalanced data

  • 期刊论文

  • 发表日期：

    2025年8月19日

查看全文
  • Xin Chen,
  • Yuejin Zhang,
  • Zhongyuan Gong,
  • Qi Shi,
  • Shuying Gong,
  • Zhuo Li,
  • Dixin Huang,
  • Nan Jiang

摘要

This research addresses network security by introducing a novel intrusion detection strategy, AGM, which combines the Adaptive Synthetic (ADASYN) sampling technique with a Gaussian Mixture Model (GMM) clustering approach. Traditional deep learning frameworks are enhanced by implementing the C3BANet model, which integrates Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Channel-Attention mechanisms. The study further investigates the impact of Principal Component Analysis (PCA) with varying principal component counts on model performance. Thorough preprocessing of the UNSW-NB15 and CICIDS2017 datasets was conducted to eliminate inconsistencies, and AGM sampling was applied to balance class representation. Our experiments show that the C3BANet model achieves exceptional detection rates of 99.99% and 96.82% in binary and multi-class settings on the UNSW-NB15 dataset, and 99.78% and 99.57% on the CICIDS2017 dataset, respectively. These results significantly surpass current leading algorithms, demonstrating both methodological innovation and practical advantages in intrusion detection.