Examining the Distribution of Keystroke Dynamics Features on Computer, Tablet and Mobile Phone Platforms

Abstract

In Keystroke Dynamics (KD) literature, the normal distribution is the first-choice probability distributions (PD) used to model KD data. A recent study has shown that the log-logistic distribution is the best distribution to model KD data that was obtained on computer keyboards. Since the type of keyboard used can affect the characteristics of the KD features, the aim of this paper is to evaluate the impact of the keyboard used for KD data acquisition on the fits of some PDs. A public KD dataset collected using a desktop, phone and tablet keyboard was used for this study. The fits of eight 2-parameters PDs were evaluated on the dataset using the Kolmogorov–Smirnov goodness-of-fit statistic. The log-logistic distribution was ranked the best in fitting the KD data and this performance was independent of the devices considered. This is important towards generation of synthetic datasets for KD research.