A Comparative Analysis of Machine Learning Models for Behavioral Biometric Authentication using Keystroke Dynamics
Adarsh Muralidharan, Amir Eaman, Esteve Hassan · Procedia Computer Science · 2025
Behavioral Biometrics provides a secure method to authenticate users in computer systems. Keystroke dynamics offers a promising approach in behavioral biometrics for user authentication in computer systems because users exhibit distinctive characteristics during typing. This study uses timing data from the Carnegie Mellon University (CMU) benchmark dataset to systematically evaluate the performance of a diverse set of machine learning models in classifying users based on their keystroke behavior. The machine learning models include traditional algorithms such as Random Forest, Support Vector Machine (SVM), and K-Nearest Neighbors (KNN), and advanced gradient boosting techniques like XGBoost, LightGBM, and deep learning architectures, specifically Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). Our results demonstrate that the LightGBM model achieves the highest accuracy of 94.68%, significantly outperforming prior hybrid approaches like the POHMM/SVM Model (86.8%). These findings contribute valuable insights for the future development of authentication applications using behavioral biometrics.