A Lightweight Approach to Continuous Authentication through Keystroke Dynamics Using Neural Networks

Authors

  • Hiwot Kibebew Author
  • Taye Abdulkadir Author

Keywords:

Machine Learning, Biometrics, Continuous authentication, Digraphs, Keystroke dynamics, Linear dependence, Neural Network, Trigraphs

Abstract

Integrating continuous authentication into computers through keystroke dynamics using machine learning/artificial intelligence demands real-time monitoring of signals from the keyboard in order to verify if the active user is legitimate. Since such process needs to run continuously, it must use minimal memory and computing power as a functional requirement. However, prior studies depend on complex high-dimensional feature engineering strategies for key-names, overlapping time-based features and heavier deep learning architectures.   

To bridge the gap, this paper proposes lightweight approaches to feature engineering and shallower architecture for continuous authentication using keystroke dynamics. A shallow architecture of hybrid 1-dimensional Convolutional Neural Networks (1D-CNN) with Long-Short Term Memory (LSTM) is implemented. The model is trained and validated using Buffalo keystroke dynamics free-text dataset achieving a low Equal Error Rate (EER). The approach provides lightweight treatment for the problem through denser vector representation of keystrokes and deeper data understanding. The main contribution of the paper is the design of compact representation of keystroke data and a shallower architecture that works on shorter keystroke sequence length, hence with reduced computational cost.

Published

2026-05-20