A Dynamic Selection Method for Touch-Based Continuous Authentication On Mobile Devices

Abstract

Touch biometric is one of the promising modalities to realise continuous authentication (CA) on mobile devices by distinguishing between touch strokes performed by the legitimate and illegitimate users. While the benefit of the scheme is promising, the effectiveness of different classification methods is not thoroughly understood. Little consideration has been given to dynamic selection of classifiers. In this paper, we proposed a dynamic selection method to deal with the security and usability needs of touch-based CA. Instead of classifying all touch samples using the same classifier, the method classifies each touch sample using the most promising classifier(s) from a pool of classifiers based on a certain measure of competence. The classifiers that achieve the highest level of competence will be selected to perform the classification task. We compared the proposed method with other dynamic selection methods, well-known single classifiers, as well as static ensemble methods. The experimental results show the potential and feasibility of the proposed method to improve the authentication performance of touch-based CA against the benchmark methods. We found that the proposed method can produce promising results with relatively low equal error rate in many scenarios of the datasets, with relatively high consistency.