A novel multi-scale network intrusion detection model with transformer.

Area of Science:

• Cybersecurity
• Network Security
• Machine Learning

Background:

• Intrusion Detection Systems (IDS) are crucial for network safety.
• Current IDS methods struggle with large datasets due to single-scale analysis, missing key features.
• This leads to degraded performance and overlooked potential threats.

Purpose of the Study:

• To propose a novel multi-scale transformer-based Intrusion Detection System (IDS) named IDS-MTran.
• To enhance intrusion detection by leveraging collaboration of multi-scale traffic features for broader pattern coverage.
• To improve the accuracy and stability of detecting various network attacks.

Main Methods:

• Utilized convolution operators with various kernels to generate multi-scale traffic features.
• Introduced Patching with Pooling (PwP) to enhance feature representation and inter-branch interaction.
• Designed a multi-scale transformer backbone for modeling diverse scale features and extracting intrusion trails.
• Developed Cross Feature Enrichment (CFE) to integrate and enrich multi-scale features for final output.

Main Results:

• The IDS-MTran model demonstrated superior performance compared to existing methods.
• Achieved accuracy exceeding 99% on three common datasets: NSL-KDD, CIC-DDoS 2019, and UNSW-NB15.
• Showcased enhanced ability to distinguish different types of network attacks effectively.

Conclusions:

• The proposed IDS-MTran model offers a more accurate and stable solution for network intrusion detection.
• Multi-scale feature analysis significantly broadens pattern coverage, addressing limitations of single-scale methods.
• The model's high accuracy and stability validate its effectiveness in real-world network security scenarios.