Scholarlink

<p>In the current complex and dynamic network environment, intrusion and attack detection have become critical security challenges, particularly in Software-defined Networking (SDN) architectures. Traditional detection methods based on statistics and machine learning often struggle to handle complex attack scenarios due to their high false positive rates and poor adaptability. This paper proposes an attack detection and mitigation model based on deep neural networks named CBDNN. The model integrates convolutional layers (CNN) to extract spatial features of network data and LSTM gate functions to capture temporal features, thereby significantly improving the model’s adaptability and recognition capabilities for complex attack patterns. The model exhibits strong prediction performance and adaptability by reducing the dependence on feature engineering. The InSDN dataset was used as the primary training set to evaluate the model’s performance, supplemented by cross-environment validation with datasets such as KDD Cup 1999, CIC-IDS2017, and CIC-DDoS2019. Experimental results show that the CBDNN model surpasses existing mainstream methods in accuracy and adaptability. It achieves a detection accuracy of 99.98% on the InSDN dataset and in SDN environments, significantly outperforming traditional methods. With its efficient adaptability to various traffic patterns and attack types, CBDNN provides robust support for developing next-generation universal Intrusion Detection Systems (IDS) and lays a solid foundation for practical applications.</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

CBDNN: An efficient and generalized intrusion detection method in SDN environments

  • 期刊论文

  • 发表日期：

    2026年5月16日

查看全文
  • Yufeng Jia,
  • Jiadong Ren,
  • Haitao He,
  • Xianshan Li,
  • Ke Xu

摘要

In the current complex and dynamic network environment, intrusion and attack detection have become critical security challenges, particularly in Software-defined Networking (SDN) architectures. Traditional detection methods based on statistics and machine learning often struggle to handle complex attack scenarios due to their high false positive rates and poor adaptability. This paper proposes an attack detection and mitigation model based on deep neural networks named CBDNN. The model integrates convolutional layers (CNN) to extract spatial features of network data and LSTM gate functions to capture temporal features, thereby significantly improving the model’s adaptability and recognition capabilities for complex attack patterns. The model exhibits strong prediction performance and adaptability by reducing the dependence on feature engineering. The InSDN dataset was used as the primary training set to evaluate the model’s performance, supplemented by cross-environment validation with datasets such as KDD Cup 1999, CIC-IDS2017, and CIC-DDoS2019. Experimental results show that the CBDNN model surpasses existing mainstream methods in accuracy and adaptability. It achieves a detection accuracy of 99.98% on the InSDN dataset and in SDN environments, significantly outperforming traditional methods. With its efficient adaptability to various traffic patterns and attack types, CBDNN provides robust support for developing next-generation universal Intrusion Detection Systems (IDS) and lays a solid foundation for practical applications.