Date of Award

2025

Publication Type

Thesis

Degree Name

M.Sc.

Department

Computer Science

Supervisor

Saeed Samet

Rights

info:eu-repo/semantics/openAccess

Abstract

The adoption of Virtual Reality (VR) and Augmented Reality (AR) technologies is rapidly expanding, raising new challenges in ensuring secure user authentication within immersive environments. This thesis presents a novel framework for continuous authentication using Transformer Encoder architecture, specifically tailored for eye-tracking as behavioural biometrics. Our approach is compared with the EKYT architecture and XGBoost model to provide a comprehensive evaluation. The study also evaluates the performance of a computationally efficient Machine Learning algorithms such as XGBoost model against these neural network architectures. Through a series of experiments, this research investigates the viability of eye tracking biometrics for user authentication, focusing on both monocular and binocular eye movements. Results demonstrate that binocular data, capturing vergence movements, enhances classification accuracy compared to monocular data. Transformer and DenseNet models achieved high accuracies across tasks, with the VRG task exhibiting superior performance due to its ability to capture detailed eye movement patterns. However, long-term evaluation revealed challenges in maintaining accuracy due to the dynamic nature of user gaze behavior over time. This work highlights the potential of gaze-based authentication as a reliable biometric modality and underscores the importance of adaptive learning models to address temporal variability in behavioral data

Share

COinS