Improving versatility in keystroke dynamic systems

Abstract

Keystroke dynamics is a biometric technique to identify users based on analyzing habitual rhythm patterns in the way they type. In order to implement this technique di erent algorithms to di erentiate an impostor from an authorized user were suggested. One of the most pre- cise method is the Mahalanobis distance which requires to calculate the covariance matrix with all that this implies: time processing and track each individual keystroke event. The hypothesis of this research was to nd an algorithm as good as Mahalanobis which does not require track every single keystroke event and improve, where possible, the process- ing time. To make an experimental comparison between Mahalanobis distance and euclidean normalized, a distance which only requires calcu- late the variance, an already studied dataset was used. The results were that use normalized euclidean distance is almost as good as Mahalanobis distance even in some cases could work better.