Document Type
Thesis - Open Access
Award Date
2026
Degree Name
Master of Science (MS)
Department / School
Mathematics and Statistics
First Advisor
Semhar Michael
Abstract
This study proposes a framework for continuous authentication based on individual typing patterns, using data from 75 participants in the University of Buffalo keystroke dynamics dataset. From their keystrokes, 199 measurements were extracted, capturing how long each key was held down and how quickly fingers moved between keys. A personal typing model was then built for each user from their first two sessions, and when a new block of typing arrived, the system asked a simple question: does this typing look more like the genuine user or like someone else? That question was answered by comparing the likelihood of the typing under both models, and the answer was updated block by block as more typing came in, with each block containing 200 keystrokes. The system was tested against every possible pairing of the 75 users. The results showed that the system reaches its best performance by using 3 blocks of typing which corresponds to around 600 keystrokes. At that point, the FAR is 4%, FRR is 3.53% and ERR is 3.93%. Performance stays stable and reliable from block three all the way through to block eighteen. These findings show that a person’s typing habits alone can be used to verify their identity continuously, accurately, and without any special hardware or complicated setup.
Publisher
South Dakota State University
Recommended Citation
Nartey, Matthew, "Continuous Authentication using Keystroke Dynamics via Finite Mixture Models and Likelihood Ratio" (2026). Electronic Theses and Dissertations. 2156.
https://openprairie.sdstate.edu/etd2/2156
Rights Statement
