Statistical median-based classifier model for keystroke dynamics on mobile devices

User authentication on mobile devices, smartphones, and tablets have received considerable interest lately due to the increasing dependence on these devices for storing sensitive data. Keystroke dynamics is a biometrics method of authentication that has the advantage of being hardware independent. However, enhancements are needed to make it a viable security solution on mobile devices. This paper presents a statistical median-based classifier model to serve as an anomaly detector in keystroke dynamics authentication on mobile devices. The proposed classifier uses the timing features of keystroke dynamics as well as touch features of mobile devices. Formulation and evaluation of proposed classifier have been influenced by an empirical analysis of a public dataset of keystroke dynamics on Android platform. The classifier considers the distance from the median of a feature element as an indicator of whether it relates to a genuine user or an impostor, based on previous training data of genuine users. The formula for measuring the acceptable distance to the median is derived from the distance between the minimum value of a feature element and its median. The model was evaluated through comparison with previous detectors, using the same dataset, and the results have shown a significant reduction in the equal-error-rate.