Entropy-based linking mechanisms for browser fingerprinting

Official URL: https://dx.doi.org/10.1109/BalkanCom71095.2026.11605019

Identifying visitors to a website is beneficial for achieving improved security, and more effective bot and fraud detection. While browser fingerprinting can also be used for intrusive tracking, raising legitimate privacy concerns, it remains a valuable tool when deployed for benign purposes such as authentication and fraud prevention. Browser fingerprinting is a relatively recent technique that enables websites to identify visitors based on a set of attributes derived from the configuration of their browser instance. Since these configurations can change over time, the fingerprints previously used to identify a user may become outdated. However, if a changed fingerprint can be detected as an evolved version of a previously observed one, the two can be linked, allowing the user to be recognized once again. Accurately linking evolved fingerprints leads to improved tracking time, meaning that visitors can be tracked over a longer period of time. In this paper, multiple heuristic mechanisms are proposed for use in browser fingerprint linking. These mechanisms were developed by computing the entropies and pairwise joint entropies of the fingerprint attributes in a volunteer dataset collected over a 2.5-month period and incorporating them numerically into the linking heuristics. The proposed heuristics achieve average tracking times comparable to or exceeding that of ThresholdFP, a recent heuristic method for fingerprint linking, with the best-performing trade-off-optimal variant achieving 52.1 days with 80.8% precision. Although precision can be improved through threshold selection, with a precision-optimal variant achieving 98.6% and approaching ThresholdFP's 99.5%, the heuristics yield lower precision in their tracking time-optimal and trade-off-optimal configurations. These results suggest that entropy-based scoring is a promising and previously unexplored direction for fingerprint linking, though further work is needed to simultaneously achieve high precision and long tracking durations.