Behavioral Intruder Detection Based on Browsing Patterns with Automated Grouping of Requested Webpages

Area of Science:

• Computer Science
• Cybersecurity
• Machine Learning

Background:

• Online fraud and impersonation attacks pose significant threats to digital services, particularly in the banking sector.
• Current authentication methods often lack robust behavioral analysis, relying heavily on traditional factors or limited mouse/keyboard dynamics.

Purpose of the Study:

• To investigate the efficacy of using standard web-server logs for behavioral impersonation detection in online banking.
• To develop and evaluate a scalable machine learning approach for identifying fraudulent user sessions.

Main Methods:

• Extraction and analysis of behavioral patterns from web server logs.
• Implementation of a Siamese neural network for classifying user web sessions.
• Development of an automated procedure using low-rank approximation for grouping web pages to handle variability.

Main Results:

• The Siamese neural network achieved 90% accuracy in classifying user sessions.
• The automated page grouping procedure improved classification accuracy and reduced manual data analysis.
• The methods demonstrated applicability in a banking scenario using real-world weblogs.

Conclusions:

• Behavioral patterns from web server logs can be effectively utilized for fraud detection in online services.
• The proposed automated approach offers a scalable and viable solution for combating behavioral impersonation, enhancing security in online banking.