Reliable hierarchical operating system fingerprinting via conformal prediction

Area of Science:

• Cybersecurity
• Machine Learning
• Network Security

Background:

• Operating System (OS) fingerprinting is vital for network security.
• Existing methods lack formal uncertainty quantification.
• Conformal Prediction (CP) offers guaranteed coverage but ignores OS taxonomy.

Purpose of the Study:

• To develop structured Conformal Prediction (CP) strategies for OS fingerprinting.
• To address limitations of flat CP application in taxonomic classification.
• To evaluate trade-offs between efficiency and structural consistency.

Main Methods:

• Introduced level-wise CP (L-CP) for independent hierarchical level calibration.
• Developed projection-based CP (P-CP) for upward projection of leaf-level sets.
• Evaluated both methods against validity guarantees and performance metrics.

Main Results:

• Both L-CP and P-CP satisfy validity guarantees for OS fingerprinting.
• L-CP provides tighter sets for forensics but has taxonomic inconsistencies.
• P-CP ensures consistent, nested sets for automation but is less efficient.

Conclusions:

• Structured CP offers improved OS fingerprinting with uncertainty quantification.
• A trade-off exists between L-CP's efficiency and P-CP's structural consistency.
• The choice between L-CP and P-CP depends on specific network security application needs.