Scholarlink

Rapid growth of Internet of Things technology creates substantial cybersecurity difficulties that impact intrusion detection systems that function within IoT networks. Signature-based detection through traditional IDS fails to handle modern IoT threats effectively because they cannot identify the continuous evolution of complex threats found in contemporary IoT networks. A new deep learning-based IDS hybrid solution appears in this research as a detection system that improves recognition of known and unknown threats in IoT networks. The combined framework integrates a Constrained Twin Variational Autoencoder (CT-VAE) component for feature extraction while using a Convolutional Block Attention Module (CBAM) Attention-based Transformer-BiGRU architecture to perform anomaly detection and classification. The CT-VAE method decreases the number of dimensions in IoT network traffic data to deliver better intrusion detection through effective feature conservation. A Lyre Bird Optimization (LBO) algorithm optimizes model performance through its application for hyperparameter tuning. Testing confirmed the success of the proposed system because it reached 98.29% accuracy with precision at 98.45% and recall at 98.1% while surpassing other current IDS models in the market. The new method tackles IoT network security through a highly effective solution that offers reliable performance and scalability for real-time detection.

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

Secure and Efficient Intrusion Detection for IoT Networks Using Hybrid Deep Learning Models with Constrained Twin Variational Auto-Encoder and Transformer-BiGRU Architecture

  • 章节

  • 发表日期：

    2026年

查看全文
  • Sharadha Kodadi,
  • Durga Praveen Deevi,
  • Naga Sushma Allur,
  • Koteswararao Dondapati,
  • Himabindu Chetlapalli,
  • Thinagaran Perumal

摘要

Rapid growth of Internet of Things technology creates substantial cybersecurity difficulties that impact intrusion detection systems that function within IoT networks. Signature-based detection through traditional IDS fails to handle modern IoT threats effectively because they cannot identify the continuous evolution of complex threats found in contemporary IoT networks. A new deep learning-based IDS hybrid solution appears in this research as a detection system that improves recognition of known and unknown threats in IoT networks. The combined framework integrates a Constrained Twin Variational Autoencoder (CT-VAE) component for feature extraction while using a Convolutional Block Attention Module (CBAM) Attention-based Transformer-BiGRU architecture to perform anomaly detection and classification. The CT-VAE method decreases the number of dimensions in IoT network traffic data to deliver better intrusion detection through effective feature conservation. A Lyre Bird Optimization (LBO) algorithm optimizes model performance through its application for hyperparameter tuning. Testing confirmed the success of the proposed system because it reached 98.29% accuracy with precision at 98.45% and recall at 98.1% while surpassing other current IDS models in the market. The new method tackles IoT network security through a highly effective solution that offers reliable performance and scalability for real-time detection.