Device Classification via Passive Fingerprints with Clustering Algorithm

In browser fingerprinting, device identification becomes increasingly difficult as the number of accesses increases because the number of combinations to be predicted increases exponentially. This paper proposes a method that employs a clustering algorithm to minimize the number of prediction pairs. The proposed method also evaluates whether two access logs were sourced from the same device and groups the results using the union-find algorithm for each device. The proposed method was applied to the access logs of an actual website, and experiments were conducted to investigate the impact of the pair reduction process. To investigate the impact of the reduction, the proposed method and the proposed method excluding the reduction were applied to a dataset of about 100,000 access logs. As a result, the processing time of the proposed method was reduced to about 1% compared to the proposed method without the reduction, and the adjusted Rand index improved by about 0.30. The experimental results demonstrate the feasibility of rapidly grouping devices, even when the target data contains unlearned variables or patterns.