MSc.Thesis Defense: Elif Ecem ŞAMLIOĞLU, Design and Implementation of a Threshold-Based Linking Scheme to Extend Browser Fingerprint Lifespan
Design and Implementation of a Threshold-Based
Linking Scheme to Extend Browser Fingerprint
Lifespan
Elif Ecem ŞAMLIOĞLU
Computer Science and Engineering, MSc. Thesis, 2025
Thesis Jury
Prof. Albert Levi (Thesis Advisor),
Prof. Cemal Yılmaz,
Assoc. Prof. Kübra KALKAN ÇAKMAKÇI
Date & Time: July 9th, 2025 – 1:00 PM
Place: FENS L065
Keywords: browser fingerprinting, user identification, authentication, privacy
Abstract
Browser fingerprinting is a powerful tool for user identification in financial and other security-critical applications that require strong authentication. However, due to the instability of browser attributes, fingerprints often change rapidly, reducing their lifespan and negatively impacting user convenience. We propose ThresholdFP, a novel linking algorithm designed to extend the durability of browser fingerprints without compromising precision. Instead of replacing the fingerprint after even a small change, ThresholdFP computes a difference score between fingerprints. If the total change remains below a configurable threshold, the new fingerprint is linked to its predecessor, forming a lineage over time. This allows minor attribute fluctuations to be tolerated, improving persistence without triggering additional identity checks such as CAPTCHAs. To ensure realistic performance evaluation, we collected and utilized two real-world datasets: one institutional dataset with over 235,000 fingerprints and a labeled volunteer dataset containing 5,774 fingerprints from known browser instances. The best-performing variant of ThresholdFP achieved an average tracking duration (i.e. fingerprint lifetime) of 55.7 days on the institutional dataset and 50.1 days on the volunteer dataset, while maintaining over 98% estimated precision and 99.5% actual precision, respectively. Compared to rival methods in the literature, ThresholdFP improves average fingerprint tracking duration by 24.33% to 106.30%, demonstrating its effectiveness and robustness under real-world conditions.