Cross-environment zero shot website fingerprinting
Wen Li, Xiaorong Zhang, Q. P. Ji · 2025
Website Fingerprinting is a key method for monitoring and managing networks. Current models for recognizing encrypted traffic behavior rely on labeled data and work well only when the training and testing environments match. However, realworld traffic often varies due to different environments, leading to feature drift and lower accuracy. Our work introduces a method called Cross-environment Zero-shot Website Fingerprinting (CEZS-WFP) which improves accuracy by mining and transferring traffic features from the original environment. In experiments involving 600GB of encrypted traffic across 12 different environments, CEZS-WFP increases the F1 score for access behavior recognition by 4% to 45% compared to traditional algorithms.