Analysis and evaluation of keystroke dynamics as a feature of contextual authentication

Keystroke Dynamics is a behavioural-biometrics solution that helps us to examine individuals' keystroke behaviour to decide legitimacy of the user; however, the drawback of this approach is that it has relatively high false negative and high false positive rates. There are some other anomaly detection approaches which examine more static properties like user's contextual details such as IP address, screensize, browser type etc. to detect legitimacy of the user but these approaches also suffer from false alerts. In this study, we compare different anomaly detection approaches and observe performance improvements when we combine these solutions. We first built keystroke dynamics and session context anomaly components, separately. Then, we examined how to combine the results of these machine learning components. Our experiments showed that while using weighted average ensemble model from these components improved performance, another approach which was to include keystroke dynamics scores in session context anomaly component as a new feature gives the opportunity to capture not only the keystroke dynamics scores but also the deviations of these scores and thus yields better performance