Scholarlink

<p>Every day, there are more and more intrusions, which result in numerous privacy violations, financial loss, and unauthorized information transfers, etc. While the hackers may concentrate on stealing private information, trade secrets, and confidential data to be sold to third parties for illicit profit, each type of intrusion has a specific task in mind. Limitations such as data lag on actual attacks and financial damages are caused by the false detection of assaults in the security and modifying environmental fields. Automatic abnormality detection techniques are needed to secure the necessary computing power and analyze the attacks. Therefore, this research paper proposes an effective automated detection system for intrusions utilizing a novel enhanced methodology. To address the low minority attack detection rate caused by the unbalanced training data, we employ Conditional Wasserstein GAN- Gradient Penalty (CWGAN-GP) technology to increase the minority samples. Next, the temporal features are retrieved using Res4Net, and the spatial characteristics are extracted using the Deep Concatenate Attention Augmented Convolution model (Deep CAC). With Res4Net efficiently modelling temporal dependencies and Deep CAC capturing intricate spatial patterns, this dual approach enables more accurate feature extraction. The Novel Quantized Salp Swarm Algorithm (NQSSA) is used to choose the key features. Finally, we employed a Novel Enhanced DL-based ResNet-fused External Attention Network (EResfEANet) for classifying attack categories, achieving superior categorization efficiency. Additionally, to further improve classification performance, hyperparameters were tuned using the Gooseneck Barnacle Optimization Algorithm (GBOA). The results from experimental verification on the proposed data sets demonstrate that applying oversampling techniques can enhance the detection rate of minority samples, increasing the overall accuracy rate. The proposed approach outperforms other current methods in that it requires less time for training and detection, with accuracy reaching above 99.86% in each of the three test sets.</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

An advanced attack categorization in NIDS using enhanced deep learning based EOptResfEANet methodology with hybrid feature extraction

  • 期刊论文

  • 发表日期：

    2025年4月28日

查看全文
  • Vinolia Alexander Moudiappa,
  • Kanya Nataraj,
  • Veeramalai Natarajan Rajavarman

摘要

Every day, there are more and more intrusions, which result in numerous privacy violations, financial loss, and unauthorized information transfers, etc. While the hackers may concentrate on stealing private information, trade secrets, and confidential data to be sold to third parties for illicit profit, each type of intrusion has a specific task in mind. Limitations such as data lag on actual attacks and financial damages are caused by the false detection of assaults in the security and modifying environmental fields. Automatic abnormality detection techniques are needed to secure the necessary computing power and analyze the attacks. Therefore, this research paper proposes an effective automated detection system for intrusions utilizing a novel enhanced methodology. To address the low minority attack detection rate caused by the unbalanced training data, we employ Conditional Wasserstein GAN- Gradient Penalty (CWGAN-GP) technology to increase the minority samples. Next, the temporal features are retrieved using Res4Net, and the spatial characteristics are extracted using the Deep Concatenate Attention Augmented Convolution model (Deep CAC). With Res4Net efficiently modelling temporal dependencies and Deep CAC capturing intricate spatial patterns, this dual approach enables more accurate feature extraction. The Novel Quantized Salp Swarm Algorithm (NQSSA) is used to choose the key features. Finally, we employed a Novel Enhanced DL-based ResNet-fused External Attention Network (EResfEANet) for classifying attack categories, achieving superior categorization efficiency. Additionally, to further improve classification performance, hyperparameters were tuned using the Gooseneck Barnacle Optimization Algorithm (GBOA). The results from experimental verification on the proposed data sets demonstrate that applying oversampling techniques can enhance the detection rate of minority samples, increasing the overall accuracy rate. The proposed approach outperforms other current methods in that it requires less time for training and detection, with accuracy reaching above 99.86% in each of the three test sets.