Publication Date

Spring 2023

Degree Type

Master's Project

Degree Name

Master of Science (MS)


Computer Science

First Advisor

Katerina Potika

Second Advisor

Mark Stamp

Third Advisor

Thomas Austin


biometric, keystroke authentication, deep learning, GANS


Leveraging machine learning for biometric authentication is an area of research that has seen a lot of progress within the past decade. Keystroke authentication based on machine and deep learning classifiers aims to develop a robust model that can distinguish a user from an adversary based on typing metrics (keystrokes). While keystroke authentication started with static text, where people type the same data, the shift has been to dynamic data where every user’s data varies. Recent literature has shown that with enough data, deep learning classifiers have the capacity to authenticate users with a low Equal Error Rate (EER).

However, popular deep learning classifiers are bottlenecked by the large amounts of data needed to make it efficient. This work solves the problem of the data bottleneck by utilizing Generative Adversarial Networks (GANs) to generate keystroke data with a valid label. Furthermore, the synthetic data produced by the GANS are used to train the Convolutional Neural Networks (CNN), attempting to push the EER rate even lower and resolve the data bottleneck.