A novel approach for APT attack detection based on feature intelligent extraction and representation learning.

Area of Science:

• Cybersecurity
• Network Security
• Machine Learning

Background:

• Advanced Persistent Threat (APT) attacks pose significant risks to critical organizations.
• Early detection and warning systems for APT campaigns are crucial for modern cybersecurity.

Purpose of the Study:

• To propose a novel approach for APT attack detection using Feature Intelligent Extraction (FIE) and Representation Learning (RL).
• To enhance the accuracy and efficiency of identifying APT campaigns in network traffic.

Main Methods:

• Feature Intelligent Extraction (FIE) combines Bidirectional Long Short-Term Memory (BiLSTM) and Attention networks to extract unusual behaviors from network traffic.
• Representation Learning (RL) utilizes data rebalancing and contrastive learning to optimize the classification of APT and normal IP addresses.
• The integrated Feature Intelligent Extraction and Representation Learning (FIERL) model is a novel approach.

Main Results:

• The proposed FIERL model demonstrates superior efficiency in APT attack detection.
• Experimental results show an improvement of over 5% across all measurements compared to existing studies.
• The method proves effective and reliable for identifying APT activities.

Conclusions:

• The FIERL model offers a significant advancement in APT detection capabilities.
• The combination of FIE and RL techniques provides a robust framework for cybersecurity threat intelligence.
• This research contributes a novel and effective solution to the challenge of detecting sophisticated cyber threats.