🐭 mAuth Mouse Machine Learning

Conference Paper: ConferencePaper.pdf

This project involves the development of a proof of concept model using computer mouse data that could potentially be used in an authentication scheme. The model uses KNN, SVM (SVC), and decision tree algorithms to analyze data such as mouse position, time of the event, and button press duration. The goal is to determine if this data can be used to accurately identify a user and differentiate them from potential imposters. By exploring the potential of mouse data for authentication purposes, this project aims to contribute to the growing field of biometric authentication research.

💾 Data collection

• The raw dataset is located in the ./raw_data/ folder
  • The same subjects with their extracted features are in the ./synth_data/extracted_features_data/ folder
• To extract the features, the raw data needs to be processed from preprocessing.py.

Data collection is done within collection.py, which utilizes the pynput library to collect mouse input data. Specifically, we collected the UNIX timestamp of the event, X and Y positions of the event, which button was pressed, and how long it was pressed for.

🔍 The Data

Here is a sample piece of data:

ID	Timestamp	X	Y	Button	Duration
2	1676925152.344601	901	488	-1	-1
2	1676925152.3605819	915	482	-1	-1
2	1676925157.8644285	975	301	-1	-1
2	1676925158.0728395	975	301	1	-1
2	1676925158.1127932	975	301	1	0.03995370864868164
2	1676925152.368573	924	480	-1	-1
...	...	...	...	...	...

• ID: Subject ID
• Timestamp: UNIX Timestamp of the event
• X, Y: The X and Y positions of the event, where it happened on the screen
• Button: The button that was pressed, for this, we used the following mapping:
  • 0: Left click
  • 1: Right Click
  • 2: Middle Click
  • 3: Scrolling Down
  • 4: Scrolling Up
  • 5: Button X1 (rarely used)
  • 6: Button X2 (rarely used)
• Duration: The duration of the event, only used to track how long a button press event was

Button X1 and X2 were the forward/backwards buttons on mice. This wasn't useful for the game during data collection, so will likely be cleaned out/not useful.

👀 Visualization

To visualize the data, we created a plotting script to view the mouse locations.

👪 The Team

• Team Lead - Braxton Fair
• Game Analysis and Evaluation Lead - Mahlet Asmare
• Model and Technical Paper Lead - Cole Harp and Mohammed Ahnaf Khalil