Scholarlink

<p>In order to improve the shortcomings of traditional algorithms in intrusion detection, an intrusion detection model integrating multiple networks was studied. The sparse coefficient and sparse penalty factor in the SSAE model were optimized by using the beetle whisker algorithm to form the BAS-SSAE model. The improved SSAE feature extraction mechanism was mixed with the CNN-BiGRU-Att model, and the Focal Loss function was introduced to replace the traditional loss function to deal with the imbalance problem of the data set. Through the verification of the UNSW-NB15 and NSL-KDD benchmark data sets, the model in this study can effectively improve the algorithm’s detection of various types of intrusion data. In terms of the recognition accuracy of the UNSW-NB15 data set, compared with the 94.25% of the Decisive Tree model, the 90.84% of the Logistic Regression model, the 91.57% of the Naïve Bayes model and the 92.18% of the SVM model, the performance of the model in this study is also the highest among all models, reaching 96.34%. The effectiveness of the model in this study has been proved, and it can be used to deal with related problems.</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

Network intrusion anomaly detection based on BAS-SSAE and CNN-BiGRU-attention fusion model

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月24日

查看全文
  • Nan Li,
  • Yu Wang,
  • Haibo Zhang,
  • Zhiqiang Li,
  • Weina Zhao

摘要

In order to improve the shortcomings of traditional algorithms in intrusion detection, an intrusion detection model integrating multiple networks was studied. The sparse coefficient and sparse penalty factor in the SSAE model were optimized by using the beetle whisker algorithm to form the BAS-SSAE model. The improved SSAE feature extraction mechanism was mixed with the CNN-BiGRU-Att model, and the Focal Loss function was introduced to replace the traditional loss function to deal with the imbalance problem of the data set. Through the verification of the UNSW-NB15 and NSL-KDD benchmark data sets, the model in this study can effectively improve the algorithm’s detection of various types of intrusion data. In terms of the recognition accuracy of the UNSW-NB15 data set, compared with the 94.25% of the Decisive Tree model, the 90.84% of the Logistic Regression model, the 91.57% of the Naïve Bayes model and the 92.18% of the SVM model, the performance of the model in this study is also the highest among all models, reaching 96.34%. The effectiveness of the model in this study has been proved, and it can be used to deal with related problems.