Touch-screen Behavioural Biometrics on Mobile Devices

Robust user verification on mobile devices is one of the top priorities globally from a financial security and privacy viewpoint and has led to biometric verification complementing or replacing PIN and password methods. Research has shown that behavioural biometric methods, with their promise of improved security due to inimitable nature and the lure of unintrusive, implicit, continuous verification, could define the future of privacy and cyber security in an increasingly mobile world. Considering the real-life nature of problems relating to mobility, this study aims to determine the impact of user interaction factors that affect verification performance and usability for behavioural biometric modalities on mobile devices. Building on existing work on biometric performance assessments, it asks: To what extent does the biometric performance remain stable when faced with movements or change of environment, over time and other device related factors influencing usage of mobile devices in real-life applications? Further it seeks to provide answers to: What could further improve the performance for behavioural biometric modalities?

Based on a review of the literature, a series of experiments were executed to collect a dataset consisting of touch dynamics based behavioural data mirroring various real-life usage scenarios of a mobile device. Responses were analysed using various uni-modal and multi-modal frameworks. Analysis demonstrated that existing verification methods using touch modalities of swipes, signatures and keystroke dynamics adapt poorly when faced with a variety of usage scenarios and have challenges related to time persistence. The results indicate that a multi-modal solution does have a positive impact towards improving the verification performance. On this basis, it is recommended to explore alternatives in the form of dynamic, variable thresholds and smarter template selection strategy which hold promise. We believe that the evaluation results presented in this thesis will streamline development of future solutions for improving the security of behavioural-based modalities on mobile biometrics.