This reseach project explores the offense and defense of a website fingerprinting attack. To replicate the Website Fingerprinting attack, we used Machine Learning models that analyse encrypted TLS or HTTPS packets. The most suitable model for this attack are the Tranformer model (CNN) but a KKN model (correct if wrong) was used as a baseline. also a random forest model was also used as a baseline (tbd if we keep it). The defense involves packet padding, respectively static, random and transformer padding.
Following a first analysis of the padding performance,(to be discussed in the future) this project explores the advantages padding offer, the risk reduction an attack and a prediction of the needed ressources for such operations Padding is the first recourse when it comes to website finderprinting attacks but often times the needed ressources (computing ressources, padding so called "iterations") are unpredictable.
tor_100w_2500tr = 100 website 2500 traces each from tor, closed world meaning only known websites (websites are classified with each). Raw traffic.(KU Leuven) tor_200w_2500tr = same as above but 200 websites instead of 100. Raw traffic.(KU Leuven) tor_open_200w_2000tr = 200 websites 2000 traces each from tor, open world meaning some websites not known. Raw traffic. (KU Leuven)
Source: https://github.com/DistriNet/DLWF (500w, 900w also availible)
cs_buflo = already paddded traffic
CW.npz = 95 websites, 105k total traces. 1 tap open (check the link if you want 2-5 tabs open at a time). Closed world (meaning it must trace from any of the 95 classified websites) Source: (Tsinghua university) https://github.com/Xinhao-Deng/Website-Fingerprinting-Library?tab=readme-ov-file
cd existing_repo
git remote add origin https://gitlab.maastrichtuniversity.nl/bcs_17_2740a/bcs-security-project.git
git branch -M main
git push -uf origin main
Chirila Lorenzo (i6377270) Bianca Popescu (i6378462) Amirali Faghih (i6397660) Patrick Groot (i6394949) Long Nguyen (i6389825) Derrick Ojo-Igbinoba (i6377721)