A novel distance-based algorithm for multi-user classification in keystroke dynamics

In this paper, we propose a novel distance-based localization algorithm for multi-user classification in keystroke biometrics. While this method can be applied across identification scenarios, here, we address a use-case to mitigate problems such as Facebook access misuse in shared settings. Our approach combines distance-based metric evaluation, dimensionality reduction, and localization, and is effective in dealing with challenges associated with keystroke dynamics, such as feature interaction, scale variations, and outliers. Our algorithm is evaluated with the CMU keystroke dynamics benchmark dataset and is shown to outperform classical approaches such as PCA and Kernel PCA combined with nearest neighbor classification.