Scholarlink

<p>Software-Defined Networks offer scalable, high-speed data transmission for IoT and 5&#xa0;G ecosystems but face cross-layer vulnerabilities due to dynamic infrastructure changes. Present intrusion detection system that utilizes deep learning struggles with imbalanced datasets, zero-day attack detection, and feature extraction inefficiencies in heterogeneous SDN-IoT environments. This study introduces a novel Transformer-based Multi-head Self-Attention Module (TMSAM) framework, integrating three key components: (i) to improve detection accuracy of Cross-Layer Feature Aggregation and reduce false positives in large-scale networks. (ii) To mitigate class imbalance and enhance the system’s generalization capacity using an integrated sampling technique called Tomek Borderline-Synthetic Minority Oversampling Technique (TB-SMOTE) for training samples, and (iii) Attention-Driven Transfer Learning (ATL) for swift model adaptation. Experiments on the InSDN and SDNFlow datasets show that TMSAM achieves 99.41% detection accuracy, with TB-SMOTE improving generalization by 0.3%, ATL reducing training time by 62% and quantization reduced the memory usage by <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7925_Article_IEq1.gif" Format="GIF" Height="6" Rendition="HTML" Resolution="72" Type="Linedraw" Width="17" /> </InlineMediaObject> <EquationSource Format="TEX">\(\sim\)</EquationSource> <EquationSource Format="MATHML"><math> <mo>∼</mo> </math></EquationSource> </InlineEquation> 40% reduction. The computational demands of Transformer architectures and large-scale flow analysis necessitate high-performance computing with parallel and distributed processing. By leveraging these capabilities, TMSAM supports real-time, scalable intrusion detection in 5&#xa0;G-IoT networks, offering improved resilience against polymorphic and adversarial attacks.</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

Transformer-driven security framework for SDN-IoT networks: integrating multi-headed self-attention with TB-SMOTE and attention-driven transfer learning

  • 期刊论文

  • 发表日期：

    2025年10月14日

查看全文
  • J. Benitha Christinal,
  • A . Ameelia Roseline

摘要

Software-Defined Networks offer scalable, high-speed data transmission for IoT and 5 G ecosystems but face cross-layer vulnerabilities due to dynamic infrastructure changes. Present intrusion detection system that utilizes deep learning struggles with imbalanced datasets, zero-day attack detection, and feature extraction inefficiencies in heterogeneous SDN-IoT environments. This study introduces a novel Transformer-based Multi-head Self-Attention Module (TMSAM) framework, integrating three key components: (i) to improve detection accuracy of Cross-Layer Feature Aggregation and reduce false positives in large-scale networks. (ii) To mitigate class imbalance and enhance the system’s generalization capacity using an integrated sampling technique called Tomek Borderline-Synthetic Minority Oversampling Technique (TB-SMOTE) for training samples, and (iii) Attention-Driven Transfer Learning (ATL) for swift model adaptation. Experiments on the InSDN and SDNFlow datasets show that TMSAM achieves 99.41% detection accuracy, with TB-SMOTE improving generalization by 0.3%, ATL reducing training time by 62% and quantization reduced the memory usage by \(\sim\) ∼ 40% reduction. The computational demands of Transformer architectures and large-scale flow analysis necessitate high-performance computing with parallel and distributed processing. By leveraging these capabilities, TMSAM supports real-time, scalable intrusion detection in 5 G-IoT networks, offering improved resilience against polymorphic and adversarial attacks.