Feature Fusion Approach on Keystroke Dynamics Efficiency Enhancement

In this paper we study the performance and effect of diverse keystroke feature combinations on keystroke dynamics authentication system by using fusion approach. First of all, four types of keystroke features are acquired from our collected dataset, later then transformed into similarity scores by using Gaussian Probability Density Function (GPD) and Direction Similarity Measure (DSM). Next, three fusion approaches are introduced to merge the scores pairing with different combinations of fusion rules. Result shows that the finest performance is obtained by the combination of both dwell time and flight time collectively. Finally, this experiment also investigates the effect of using larger dataset on recognition performance, which turns out to be rather consistent.