Scholarlink

The Onion Browser (Tor) provides anonymity to users by preventing adversaries from identifying which browser a user is visiting. Though it is a secure and anonymous browser, it is still vulnerable, and one of the vulnerabilities is the “website fingerprinting (WF)” attack. Preventing WF attacks using defensive techniques is a significant challenge nowadays. Several defensive techniques, such as WTFPad, Walkie-Talkie, BuFLo, Tamaraw, etc., have been applied to the network traces to protect against WF attacks. In this study, we proposed a WF attack model based on CNN that incorporates the defensive techniques (DT) of WTFPad and Walkie-Talkie. Through experimentation, it was observed that the proposed attack model outperformed even on the WTFPad protected dataset, achieving an accuracy of 90.11% in a closed-world setting and 85.01% in a real-world setting with a precision value of 0.99 and a recall value of 0.06, compared to Walkie-Talkie’s respective accuracy of 2 and 1%. The proposed CNN-based attack model outperforms both non-defended and defended Tor network traces. It demonstrates the precise identification of monitored and non-monitored websites.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced Website Fingerprinting Attack Model: A CNN-Based Approach with WTFPad and Walkie-Talkie Defenses

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Minal Sinha,
  • Mayank Dave

摘要

The Onion Browser (Tor) provides anonymity to users by preventing adversaries from identifying which browser a user is visiting. Though it is a secure and anonymous browser, it is still vulnerable, and one of the vulnerabilities is the “website fingerprinting (WF)” attack. Preventing WF attacks using defensive techniques is a significant challenge nowadays. Several defensive techniques, such as WTFPad, Walkie-Talkie, BuFLo, Tamaraw, etc., have been applied to the network traces to protect against WF attacks. In this study, we proposed a WF attack model based on CNN that incorporates the defensive techniques (DT) of WTFPad and Walkie-Talkie. Through experimentation, it was observed that the proposed attack model outperformed even on the WTFPad protected dataset, achieving an accuracy of 90.11% in a closed-world setting and 85.01% in a real-world setting with a precision value of 0.99 and a recall value of 0.06, compared to Walkie-Talkie’s respective accuracy of 2 and 1%. The proposed CNN-based attack model outperforms both non-defended and defended Tor network traces. It demonstrates the precise identification of monitored and non-monitored websites.