k-fingerprinting: A Robust Scalable Website Fingerprinting Technique
summary
New classifier combines random decision forest + k-NN fingerprints over 150 handcrafted features (top leaks: packet counts and early-page bursts) and is tested on Alexa sites and Tor hidden services with 100,000 unmonitored pages. Results: ~96% TPR but ~4% FPR on Alexa (many false alarms at scale); for 30 monitored hidden services it achieves 85% true positive with FPR as low as 0.02% against a 100,000-page world. Assumes precise load segmentation, no background tabs or downloads; code and data are published.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 3:02Contributions
- 4:28Features
- 6:54How does it work
- 8:45Base Rate
- 10:05Accuracy Metrics
- 11:55How kfingerprinting works
- 12:55Data collection
- 13:42Accuracy
- 15:12Alexa
- 15:45Hidden Service
- 17:20Limitations
- 19:21Conclusion
- 21:02Interview
related talks
40mAn Analysis of the Fingerprinting-Scripts and their Actors on the Internet
21mFp-Scanner: The Privacy Implications of Browser Fingerprint Inconsistencies
43mAround the Web in 80 Hours: Scalable Fingerprinting with Chromium Automation
14mBrowser Fingerprinting: Stalking With a Personal Touch
35mStealthy fingerprinting users with personalized blocking rules
43mPassive Fingerprinting of HTTP/2 Clients.
58mCatching Transparent Phish: Understanding and Detecting MITM Phishing Kits
44mNetsec is Dead(?): Modern Network Fingerprinting for...