A Dual-Perspective Self-Supervised IoT Intrusion Detection Method Based on Topology Reconstruction and Feature Perturbation
The rapid growth of the Internet of Things has driven intelligent advancements across various fields. However, the proliferation of IoT devices and diverse network interactions also introduces significant security challenges. As a critical technology for securing IoT, intrusion detection systems aim...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10902014/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The rapid growth of the Internet of Things has driven intelligent advancements across various fields. However, the proliferation of IoT devices and diverse network interactions also introduces significant security challenges. As a critical technology for securing IoT, intrusion detection systems aim to identify potential threats by analyzing network traffic features. Yet, traditional models struggle to capture the complex topological structures in IoT environments, and their training often relies heavily on large amounts of labeled data, making them unsuitable for IoT settings where massive data is continually generated. This paper presents a graph-based framework leveraging self-supervised learning to address these challenges. This framework employs a graph encoder to capture topological information and generate graph embeddings, utilizing topology reconstruction and feature perturbation to create diverse loss graphs that enhance feature learning. Additionally, the custom adaptive cosine loss function dynamically adjusts the loss weights of different samples and utilizes cosine similarity to align the embeddings of real and predicted graphs, thereby maximizing mutual information between local embeddings of real graphs and global summaries. The model was evaluated on four public datasets, and the experimental results demonstrate that the proposed method significantly improves detection performance. It outperforms existing self-supervised graph neural network models, validating the applicability of this approach in settings with unlabeled data. |
|---|---|
| ISSN: | 2169-3536 |