Scholarlink

<p>In intrusion detection systems, a critical imbalance in the attack and normal sample quantities leads to low detection accuracy for minority-class attacks. This study introduces an innovative method, namely Parallel Convolutional Stacked Autoencoder (PConVSA-Net), for minority-class attack detection. Initially, the input data are extracted from the dataset and undergo a normalization process, where log transformation is used to modify the data into a standardized form. Further, the normalized data are then passed to the feature selection module. Here, a Support Vector Machine with Recursive Feature Elimination (SVM-RFE) is employed to select the features. The weight parameters of SVM are adjusted by the introduced Taylor Pufferfish Optimization Algorithm, which is established by compounding the Taylor-series concept into the Pufferfish optimization algorithm. Later, the Synthetic Minority Oversampling Technique is utilized to perform data augmentation. Lastly, the minority-class attack detection is done by the established PConvSA-Net, which is developed by merging the Parallel Convolutional Neural Network (Parallel-CNN) and Deep Stacked Autoencoder. For dataset 1, the introduced PconvSA-Net attained high accuracy and true-positive rate of 91.876% and 94.326%, a low false-positive rate of 9.786%, precision of 90.878%, and F1-score of 92.570%.</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

SVM-RFE with optimization-based feature selection and parallel convolutional stacked autoencoder for detecting minority-class sample attacks

  • 期刊论文

  • 发表日期：

    2025年9月25日

查看全文
  • A. N. Sasikumar,
  • S. Suresh

摘要

In intrusion detection systems, a critical imbalance in the attack and normal sample quantities leads to low detection accuracy for minority-class attacks. This study introduces an innovative method, namely Parallel Convolutional Stacked Autoencoder (PConVSA-Net), for minority-class attack detection. Initially, the input data are extracted from the dataset and undergo a normalization process, where log transformation is used to modify the data into a standardized form. Further, the normalized data are then passed to the feature selection module. Here, a Support Vector Machine with Recursive Feature Elimination (SVM-RFE) is employed to select the features. The weight parameters of SVM are adjusted by the introduced Taylor Pufferfish Optimization Algorithm, which is established by compounding the Taylor-series concept into the Pufferfish optimization algorithm. Later, the Synthetic Minority Oversampling Technique is utilized to perform data augmentation. Lastly, the minority-class attack detection is done by the established PConvSA-Net, which is developed by merging the Parallel Convolutional Neural Network (Parallel-CNN) and Deep Stacked Autoencoder. For dataset 1, the introduced PconvSA-Net attained high accuracy and true-positive rate of 91.876% and 94.326%, a low false-positive rate of 9.786%, precision of 90.878%, and F1-score of 92.570%.