Keystroke-based Identification

Python code for collecting and authenticating a user based on their keystroke patterns.

Running

• Using Docker:

  docker build -t ucontacti/key-stroke:latest .  
  docker run -p 5000:5000 -it ucontacti/key-stroke
  

• Using Python(Make sure you are using python 3.6 or above):

  pip install -r requirements.txt  
  python app.py  
  

Either way, the web page is accessible from localhost:5000

Project explanation

The code consists of the following parts

Data collection

To collect data a web interface was implemented using Flask and keystrokes timing were sampled with the aid of Javascript event handlers.

Feature Extraction

feature_extractor the takes raw data and produces 3 different feature data set

Training

Training was done using two major algorithms

• Classic Machine-Learning
  • KNN
  • SVM
• Neural Network (Pytorch)

Testing

The web platform also uses the model and performs a authentication using previous mentioned methods

Results

The results of the authentication was over 90% and the detail report can be found here

Authors

• Saleh Daghigh - ucontacti
• Mahdi Paktinat

This project was done under the supervision of Dr. Michael Mock from Fraunhofer Institute.