Investigating Minimal Keystroke Samples for Reliable User Authentication
M. S. Arjun, P Aswin, Liya Treesa Philip, Ashlyn Wilson Sasthampadavil, Athul K Alias, Parvathy Krishna R, Aparna Mohanty · 2025
In today’s increasingly connected world, achieving secure yet user-friendly authentication is more critical than ever. Keystroke dynamics, which captures the unique timing patterns of individuals’ typing behavior—specifically hold and flight times—offers a promising solution. However, real-world scenarios often involve limited input, such as short passwords or brief system interactions.This study investigates the feasibility of accurate user authentication using minimal keystroke sequences. We evaluated two machine learning techniques: a supervised Random Forest classifier trained with both genuine and impostor data, and an unsupervised One-Class Support Vector Machine (SVM) model trained solely on genuine data. By progressively varying input length from 3 to 20 keystrokes, we analyzed model performance using classification accuracy and Area Under the ROC Curve (AUC). Our findings suggest that reliable authentication is achievable with as few as 5–7 keystrokes, enabling lightweight, real-time verification in applications like mobile devices and embedded systems.