Scholarlink

<p>Software-Defined Networking integrated with the Internet of Things introduces new security vulnerabilities due to the growing complexity of network traffic. To address these challenges, this paper proposes a CNN-based intrusion detection model enhanced with feature normalization to ensure balanced learning across heterogeneous traffic features. Unlike most prior inSDN studies limited to binary classification, our work focuses on multiclass intrusion detection and evaluates the model’s robustness across unseen datasets. The proposed model is tested on three benchmark datasets inSDN, IoT-SDN IDS, and SDN Intrusion Detection with no data leakage between training and testing. It achieves state-of-the-art performance, with detection accuracies of 99.98%, 99.86%, and 99.68%, respectively, consistently outperforming approaches such as WNIDS, CNN-RF, LSTM, SelectKBest-DT, and V-NKDE. Additional evaluation using precision, recall, and F1-score confirms robustness against class imbalance and resilience to overfitting. These results highlight CNN1D as a scalable and reliable solution for real-world intrusion detection in SDN-enabled IoT environments.</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

Robust CNN-Based Threat Detection in SDN-Enabled IoT Networks

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Safae Khalis,
  • Karima Hassini,
  • Mohammed Chemmakha,
  • Omar Habibi,
  • Mohamed Lazaar

摘要

Software-Defined Networking integrated with the Internet of Things introduces new security vulnerabilities due to the growing complexity of network traffic. To address these challenges, this paper proposes a CNN-based intrusion detection model enhanced with feature normalization to ensure balanced learning across heterogeneous traffic features. Unlike most prior inSDN studies limited to binary classification, our work focuses on multiclass intrusion detection and evaluates the model’s robustness across unseen datasets. The proposed model is tested on three benchmark datasets inSDN, IoT-SDN IDS, and SDN Intrusion Detection with no data leakage between training and testing. It achieves state-of-the-art performance, with detection accuracies of 99.98%, 99.86%, and 99.68%, respectively, consistently outperforming approaches such as WNIDS, CNN-RF, LSTM, SelectKBest-DT, and V-NKDE. Additional evaluation using precision, recall, and F1-score confirms robustness against class imbalance and resilience to overfitting. These results highlight CNN1D as a scalable and reliable solution for real-world intrusion detection in SDN-enabled IoT environments.