Analysis of Typing Pattern in Identifying Soft Biometric Information and Its Impact in User Recognition

As of now, the performance of keystroke dynamics biometric in user recognition is not acceptable in practice due to intra-class variations, high failure to enroll rate (FER) or various troubles in data acquisition methods or diverse use of sensing devices. As per the previous study, the performance of this technique can be improved by incorporation of gender information, a soft biometric characteristic, extracted from the typing pattern on a computer keyboard that provides some additional information about the user. This soft biometric trait has low user discriminating power but can be used to enhance the performance of user recognition in accuracy and time efficiency. Furthermore, it has been observed that the age group (18–30/30+ or <18/18+), gender (male/female), handedness (left-handed/right-handed), hand(s) used (one hand/both hands), typing skill (touch/others), and emotional states (anger/excitation) can be extracted from the way of typing on a computer keyboard for single predefined text. In this paper, we are interested in identifying multiple soft biometric traits using two leading machine learning methods: support vector machine with radial basis function (SVM-RBF) and fuzzy-rough nearest neighbor with vaguely quantified rough set (FRNN-VQRS) on multiple publicly available authentic and recognized keystroke dynamics datasets collected through a computer keyboard as well as touchscreen phone. The performance of machine learning methods are changed significantly in changing dataset in keystroke dynamics domain, but the evaluation performance of FRNN-VQRS in our experiment is promising and consistent in identifying traits. At the end, the impacts of the incorporation of soft biometric traits with primary biometric characteristics in user recognition are presented and compared the evaluation performance of nine anomaly detectors.