Convolutional Neural Networks for Dynamic Browser Fingerprint Linking

Facing the increasing complexity of cybersecurity threats, traditional defenses are proving inadequate. Browser fingerprinting offers a promising approach to user identification through analysis of device and software characteristics. However, the volatile nature of fingerprint information due to browser updates and device changes hinders its effectiveness. This paper introduces ATCN, a novel dynamic browser fingerprint linking algorithm based on Temporal Convolutional Networks (TCN). Leveraging time series analysis and an attention mechanism, ATCN improves both the accuracy and generalizability of fingerprint linking. Experimental results show that ATCN achieves a linking duration exceeding 50 days and a linking rate above 72 % for infrequent users, outperforming LSTM and random forest algorithms by 5 % and approximately 12 %, respectively. These findings highlight the potential of ATCN for enhancing cybersecurity applications.