Aspects of Continuous User Identification Based on Free Texts and Hidden Monitoring

Abstract

This paper investigates some specific features of continuous user identification based on hidden monitoring of keystroke dynamics when creating a free text. Our analysis of static identification approaches does not reveal any significant limitations on their application to continuous identification. The main feature of continuous identification is the method for collecting dynamic information about key presses and the correction of templates of registered users. The effectiveness of including additional classification features in recognition algorithms, e.g., those associated with the frequency of letters in texts, is demonstrated. A software application is developed to collect and analyze keystroke rhythm samples of users. Research in the domain of users with good computer skills shows quite satisfactory user recognition accuracy (87% on average). Moreover, the accuracy does not depend on the metric distance selected for recognition and improves with the use of scaling factors for letter frequency.