Scholarlink

<p>As the Internet of Things (IoT) advances, its network environment grows complex, heightening IoT security concerns due to frequent attacks. Intrusion detection, a key network security technology, remains essential in the IoT but has limitations. In network traffic within the IoT, the problem of data imbalance is common, normal traffic greatly exceeds abnormal traffic. This imbalance can cause deep learning models to overfit the majority class during training. To address this issue, a data augmentation model is designed that leverages the conditional generation capabilities of Conditional GAN, combined with an improved DCGAN based on convolutional layers, to generate various types of high-quality minority class data. This study introduces Conditional Deep Convolutional Generative Adversarial Network-Convolutional Neural Network (CondGAN-CNN), a data augmentation model aimed at balancing the distribution of minority classes in IoT traffic data. This model improves the authenticity and quality of data generation by introducing convolutional layers based on the GAN framework and adding BN layers to enhance network stability. Then, combining the conditional generation idea of cGAN, minority class data labels are used as generation conditions to achieve efficient multi-category data generation. CondGAN-CNN effectively balances the dataset, reducing bias toward the majority class. Compared to two common data augmentation methods, it improves accuracy and reliability for IoT intrusion detection. The model addresses IoT security needs by effectively identifying anomalous traffic and offers a strong solution for IoT security research, enhancing minority class data better than traditional methods.</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

Boosting IoT security: a novel CondGAN-CNN framework for enhanced intrusion detection in imbalanced traffic scenarios

  • 期刊论文

  • 发表日期：

    2026年6月17日

查看全文
  • Tariq Mahmood,
  • Tanzila Saba,
  • Amjad Rehman,
  • Faten S. Alamri,
  • Muhammad Nazam Maqbool

摘要

As the Internet of Things (IoT) advances, its network environment grows complex, heightening IoT security concerns due to frequent attacks. Intrusion detection, a key network security technology, remains essential in the IoT but has limitations. In network traffic within the IoT, the problem of data imbalance is common, normal traffic greatly exceeds abnormal traffic. This imbalance can cause deep learning models to overfit the majority class during training. To address this issue, a data augmentation model is designed that leverages the conditional generation capabilities of Conditional GAN, combined with an improved DCGAN based on convolutional layers, to generate various types of high-quality minority class data. This study introduces Conditional Deep Convolutional Generative Adversarial Network-Convolutional Neural Network (CondGAN-CNN), a data augmentation model aimed at balancing the distribution of minority classes in IoT traffic data. This model improves the authenticity and quality of data generation by introducing convolutional layers based on the GAN framework and adding BN layers to enhance network stability. Then, combining the conditional generation idea of cGAN, minority class data labels are used as generation conditions to achieve efficient multi-category data generation. CondGAN-CNN effectively balances the dataset, reducing bias toward the majority class. Compared to two common data augmentation methods, it improves accuracy and reliability for IoT intrusion detection. The model addresses IoT security needs by effectively identifying anomalous traffic and offers a strong solution for IoT security research, enhancing minority class data better than traditional methods.