Sign in Subscribe
Technical institutes

Industrial Internet of Things Cyber Threats Detection Through Deep Feature Learning

Strategic Insights

— 1 min read

Researchers at the Vellore Institute of Technology in India have developed a new method for detecting cyber threats in industrial Internet of Things (IoT) systems, achieving an accuracy of 99.86% and 99.62% on two different datasets. The proposed framework uses a lightweight hybrid deep learning algorithm to identify intrusion and prevent overfitting problems during training. The method involves resolving data imbalance using the Euclidean-based synthetic minority oversampling technique (EbSmoT), eliminating redundant features with the Information Gain and Fisher score-based technique (IG-FST), and obtaining higher-level feature representation using the Bi-LSTM ResNet-based convolutional autoencoder (BR-CAE). The Stacked Sparse autoencoder-based Particle Swarm Probabilistic Neural Network (SAE-PSPNN) is then used for attack detection and classification.

Key Takeaways:

  • The proposed method uses a lightweight hybrid deep learning algorithm to detect cyber threats in industrial IoT systems.
  • The method achieves an accuracy of 99.86% on the ToN_IoT dataset and 99.62% on the UNSW-NB15 dataset.
  • The data imbalance problem is resolved using the Euclidean-based synthetic minority oversampling technique (EbSmoT).
  • The Information Gain and Fisher score-based technique (IG-FST) is employed to eliminate redundant features and avoid overfitting problems during training.
  • The Bi-LSTM ResNet-based convolutional autoencoder (BR-CAE) is used to obtain higher-level feature representation.
  • The Stacked Sparse autoencoder-based Particle Swarm Probabilistic Neural Network (SAE-PSPNN) is used for attack detection and classification.
  • The method has been evaluated using two different datasets, the UNSW-NB15 dataset and the ToN_IoT dataset.
  • The research has been peer-reviewed and published in the Transactions on Emerging Telecommunications Technologies journal.
  • The authors of the research include R. Vijay Anand, G. Magesh, I. Alagiri, Madala Guru Brahmam, C. Senthil Kumar, M. Kesavan, and Azween Bin Abdullah.

Statistics:

  • 99.86% accuracy on the ToN_IoT dataset.
  • 99.62% accuracy on the UNSW-NB15 dataset.
  • The proposed method has achieved a higher accuracy compared to previous methods.
  • The data imbalance problem affects the performance of the model, but EbSmoT resolves it effectively.
  • The Bi-LSTM ResNet-based convolutional autoencoder (BR-CAE) enhances feature representation significantly.

Sources:

  • Industrial Internet of Things Cyber Threats Detection Through Deep Feature Learning and Stacked Sparse Autoencoder Based Classification. Transactions on Emerging Telecommunications Technologies, 2025;36(9).
  • Vellore Institute of Technology, Sch Comp Sci Engn & Informat Syst, Vellore, India.
  • Wiley, 111 River St, Hoboken 07030-5774, NJ, USA.
  • Transactions on Emerging Telecommunications Technologies - onlinelibrary.wiley.com/journal/10.1002/(ISSN)2161-3915.

Read more