Classifying Biometric Systems Users among the Doddington Zoo: Application to Keystroke Dynamics

Doddington zoo defines four categories of users when using a biometric system related to their difficulty to be recognized or attacked. In this paper, we propose an original work consisting in predicting for any biometric modality the associated animal in the Doddington menagerie related to a user given few captured biometric samples. Such a prediction could be useful for many applications, as for example, to adapt the behavior of biometric systems to each user. In this work, we apply this methodology to keystroke dynamics as it is an interesting behavioral biometric modality for user authentication. It consists in analyzing the way of typing of a user in order to recognize him/her. We use a significant keystroke dynamics dataset and we demonstrate through experimental results the benefit of the proposed approach.