# Traffic Analysis

**Repository Path**: xxdxxdxxd/traffic-analysis

## Basic Information

- **Project Name**: Traffic Analysis
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 1
- **Forks**: 1
- **Created**: 2023-11-11
- **Last Updated**: 2024-10-22

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Tasks & Models & Datasets

> This file discusses the appropriate tasks, models and datasets to be utilized for robustness certification.

## Website fingerprinting

### Representative models

![whiteboard_exported_image](README.assets/whiteboard_exported_image.png)

[1] Automated Website Fingerprinting through Deep Learning

[2] Deep Fingerprinting: Undermining Website Fingerprinting Defenses with Deep Learning

[3] Var-CNN: A Data-Efficient Website Fingerprinting Attack Based on Deep Learning

[4] Tik-tok: The utility of packet timing in website fingerprinting attacks

[5] Subverting Website Fingerprinting Defenses with Robust Traffic Representation

[6] Triplet Fingerprinting: More Practical and Portable Website Fingerprinting with N-shot Learning

[7] An {Input-Agnostic} Hierarchical Deep Learning Framework for Traffic Fingerprinting

[8] Effective Attacks and Provable Defenses for Website Fingerprinting

[9] Website Fingerprinting at Internet Scale

[10] k-fingerprinting: A Robust Scalable Website Fingerprinting Technique

[11] High Precision Open-World Website Fingerprinting

### Detailed analysis

#### 1. k-NN

https://www.cs.sfu.ca/~taowang/wf/index.html

- **Influence factors:** length (?), timestamp, direction.
- **Feature set:** 1000~4000 extracted features (general features, unique packet lengths, packet ordering, concentration of outgoing packets, bursts and intial packet lengths).
- **Algorithm:** k-Nearest Neighbour classifier.
- **Dataset**
  - WANG14: Tor cell traces.
    - 100 monitored websites (only index pages) with 90 instances each.
    - 9000 unmonitored websites with 1 instance each.
- **Data type****:** (Tor cell level) sequence of <timestamp, ±length>. For Tor, length is set to 1.



#### 2. CUMUL

https://www.informatik.tu-cottbus.de/~andriy/zwiebelfreunde/

- **Influence factors:** length, timestamp, direction.
- **Feature set:** n = 100 interpolants of the cumulative packet length curve.
- **Algorithm:** libSVM.
- **Dataset**
  - ALEXA100: re-recorded version of WANG13.
    - 100 monitored websites with 100 instances each.
  - RND-WWW: random samples of web pages visited by typical users.
    - 1125 foreground webpages (from 712 websites).
    - 118884 background webpages (34580 websites).
  - TOR-Exit: pages that are actually accessed through the Tor network.
    - 211148 unique webpages (each website is represented by fewer than 2,000 web pages).
  - WEBSITES: websites and their subpages.
    - 20 websites.
    - For each website, 1 index page (90 instances) + 50 subpages (15 instances each).
- **Data type****:** (TLS, or TCP level) sequence of <±length>.



#### 3. k-FP

https://github.com/jhayes14/k-FP

- **Influence factors:** length (?), timestamp, direction.
- **Feature set:** 150 extracted features (packet numbers, ordering, concentrations, etc).
- **Algorithm:** Random Forest.
- **Dataset:** Wang14 (the same with kNN), along with self-collected dataset gathered in the similar way.
- **Data type****:** (TCP/IP packet level) sequence of <timestamp, ±length>. For Tor, length is set to 1.



#### 4. openWF

https://github.com/literaltao/openwf

This work proposes a precision optimization (PO) approach and tests it on 7 models (including k-NN, CUMUL, k-FP, DF).

- **Dataset:** Wang20000
  - 100 monitored pages (200 instances each) + 80000 unmonitored pages (1 instance each).
- **General** **data type****:** (Tor cell, TLS, or TCP level) sequence of <timestamp, ±length>