Anatomy of secondary features in keystroke dynamics - achieving more with less

Abstract

Keystroke dynamics is an effective behavioral biometric for user authentication at a computer terminal. While many distinctive features have been used for the analysis of acquired user patterns and verification of users transparently, a group of features such as Shift and Comma has always been overlooked and treated as noise. In this paper, we define these normally ignored features as secondary features and investigate their effectiveness in user verification/authentication. By evaluating all the available secondary features, we have found that they contain valuable information that is characteristic of individuals. With a limited number of secondary features, we achieved a promising Equal Error Rate (EER) of 2.94% and Area Under the ROC Curve (AUC) of 0.9940 for classification on a publicly available data set. Surprisingly, this result compares well with the results obtained from primary features by other researchers and we are able to achieve quality results with fewer data records, indicating a reduced training time in comparison.