Nwokedi, Kenechukwu Chinedu (2024) Continuous User Authentication for Secure Messaging Using Advanced Machine Learning Models. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (2MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (4MB) | Preview |
Abstract
Being unable to offer continuous verification once a person logs into a system is a major drawback of conventional authentication methods such as passwords and fingerprints (Dhaka, Rao, and Chaurasia, 2022) Without performing additional security checks, applications using these traditional methods are vulnerable to attacks from unauthorized users. This is a serious vulnerability, particularly for messaging services where users might keep open sessions for a long time. So, modern applications need to meet the needs of the ever-evolving cybersecurity landscape. In this study, a keystroke and mouse dataset are utilized for continuous authentication, extracting user behavioral patterns. Features such as average dwell time, flight time, and mouse trajectory are computed from the data and used to train different Machine Learning (ML) classification algorithms such as Decision Tree Classifier, Random Forest Classifier, Support Vector Machine, and Gradient Boosting Classifier. The algorithms use the patterns of behavior to classify users as “legitimate” or “illegitimate” and their performances are further compared using metrics like accuracy and F measure. Lastly, a custom messaging application is designed, and it was discovered that a selected trained model could be incorporated into the application framework for continuous, seamless authentication. The results suggest that machine learning models are effective at classifying users and behavioral biometric data can be used by applications for more advanced security.
Actions (login required)
![]() |
View Item |