Hybrid ResNet-1D-BiGRU with Multi-Head Attention for Cyberattack Detection in Industrial IoT Environments
Summary: arXiv:2604.06481v1 Announce Type: cross
Abstract
This study introduces a hybrid deep learning model for intrusion detection in Industrial IoT (IIoT) systems. The model combines ResNet-1D, Bidirectional Gated Recurrent Units (BiGRU), and Multi-Head Attention (MHA) techniques to facilitate effective spatial-temporal feature extraction and attention-based feature weighting. The model is designed to tackle the challenge of class imbalance by applying the Synthetic Minority Over-sampling Technique (SMOTE) during the training phase on the EdgeHoTset dataset.
Model Performance
The hybrid model demonstrated impressive performance metrics, achieving an accuracy of 98.71%, with a loss of just 0.0417%. Moreover, it exhibited low inference latency of 0.0001 seconds per instance, showcasing its strong real-time capabilities. To evaluate the generalizability of the model, it was subsequently tested on the CICIoV2024 dataset, where it achieved remarkable results, including:
- Accuracy: 99.99%
- F1-score: 99.99%
- Loss: 0.0028
- False Positive Rate (FPR): 0%
- Inference Time: 0.00014 seconds per instance
Comparative Analysis
When compared to existing intrusion detection methods, the proposed hybrid model consistently outperformed its counterparts across all evaluated metrics and datasets. The robust performance indicates not only the effectiveness of the architecture in detecting cyberattacks in real-time but also its potential applicability in various Industrial IoT settings.
Conclusion
The integration of ResNet-1D, BiGRU, and Multi-Head Attention in the proposed hybrid model represents a significant advancement in the field of cybersecurity for Industrial IoT. By effectively addressing the challenges of class imbalance and demonstrating high accuracy and efficiency, this model paves the way for future research and developments in secure IIoT environments.
Future Work
Future research directions may include:
- Further optimization of the model for even lower inference times.
- Exploration of additional feature extraction techniques to enhance detection capabilities.
- Application of the model to other datasets to assess its adaptability across diverse environments.
- Investigation into real-time deployment challenges and solutions in operational IIoT systems.
In conclusion, the hybrid ResNet-1D-BiGRU with Multi-Head Attention model stands as a promising solution for enhancing cybersecurity measures in Industrial IoT, ensuring that systems remain resilient against evolving cyber threats.