Scholarlink

<p>The rapid development of 6G, Internet of Things (IoT), and critical infrastructure networks has brought unprecedented connectivity, scalability, and heterogeneity, also increasing the surface area for cyberattacks. Conventional intrusion detection systems (IDSs) have problems with the diverse variety of traffic patterns, domain changes, and adversarial activity. To address these challenges, this paper presents the Hybrid Deep Computational Intelligence Network (HDCI-Net), a single framework that integrates convolutional, recurrent, fuzzy, and adversarial learning concepts in cross-domain intrusion detection. The main aspect of this work is to generate UniSec-Fusion2025, a massive, multi-domain dataset comprising traffic of SANDI-2024, TII-SSRC-23, NF-UQ-NIDS-v2, and individual 5G/6G prototype captures. The dataset was preprocessed to make the data statistically consistent and realistic across domains. HDCI-Net uses dual branch CNN-BiLSTM encoder with multi-head attention to capture both local and temporal dependency and a Fuzzy feature weighting module is used for better interpretability and stability. The Domain Adversarial Neural Network (DANN) promotes domain-invariant feature learning so that the model can effectively generalize to unseen environments. Experimental evaluation shows strong performance with 99.42% of accuracy, a 99.36% weighted F1-score, a macro ROC-AUC of 0.9992 - but with an inference latency of only 1.7 ms per flow. Explainability analysis with SHAP showed that important aspects of the traffic descriptors (e.g., entropy of the flow, burstiness, packet rate) dominate the decision process of the model, which confirms a transparent and semantically consistent reasoning process. The UniSec-Fusion2025 + HDCI-Net framework provides a promising approach in cross-domain, interpretable, and real-time intrusion detection of heterogeneous 6G and IoT frameworks. This development brings the field a step further towards autonomous, adaptive, and reliable cybersecurity provision of next-generation digital infrastructures.</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

HDCI Net for interpretable cross domain intrusion detection in 6G IoT and critical infrastructure networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月28日

查看全文
  • Amjad Qashlan

摘要

The rapid development of 6G, Internet of Things (IoT), and critical infrastructure networks has brought unprecedented connectivity, scalability, and heterogeneity, also increasing the surface area for cyberattacks. Conventional intrusion detection systems (IDSs) have problems with the diverse variety of traffic patterns, domain changes, and adversarial activity. To address these challenges, this paper presents the Hybrid Deep Computational Intelligence Network (HDCI-Net), a single framework that integrates convolutional, recurrent, fuzzy, and adversarial learning concepts in cross-domain intrusion detection. The main aspect of this work is to generate UniSec-Fusion2025, a massive, multi-domain dataset comprising traffic of SANDI-2024, TII-SSRC-23, NF-UQ-NIDS-v2, and individual 5G/6G prototype captures. The dataset was preprocessed to make the data statistically consistent and realistic across domains. HDCI-Net uses dual branch CNN-BiLSTM encoder with multi-head attention to capture both local and temporal dependency and a Fuzzy feature weighting module is used for better interpretability and stability. The Domain Adversarial Neural Network (DANN) promotes domain-invariant feature learning so that the model can effectively generalize to unseen environments. Experimental evaluation shows strong performance with 99.42% of accuracy, a 99.36% weighted F1-score, a macro ROC-AUC of 0.9992 - but with an inference latency of only 1.7 ms per flow. Explainability analysis with SHAP showed that important aspects of the traffic descriptors (e.g., entropy of the flow, burstiness, packet rate) dominate the decision process of the model, which confirms a transparent and semantically consistent reasoning process. The UniSec-Fusion2025 + HDCI-Net framework provides a promising approach in cross-domain, interpretable, and real-time intrusion detection of heterogeneous 6G and IoT frameworks. This development brings the field a step further towards autonomous, adaptive, and reliable cybersecurity provision of next-generation digital infrastructures.