A flexible approach for biometric menagerie on user classification of keystroke data

Biometric systems aim to provide reliable authentication and verification of users. The behaviour of the users may alter the authentication performance when accessing these systems. Therefore, clustering users based on their actions is crucial. A biometric menagerie defines and labels user groups statistically according to their variability. However, determining groups is a fuzzy process and it may lead to inconsistencies. In this work, a novel and flexible approach is introduced based on the classification performance of the users data collected in a database without imposing any other restrictions. According to the performance measures obtained from the confusion matrix of the classification algorithms, users are ranked and then clustered. Additionally, the norm of a confusion matrix is offered augmenting the state-of-the-art performance metrics. The proposed scheme is evaluated using the behavioural biometrics modality on two benchmark keystroke databases. The performance results successfully illustrate the alternative way of grouping and identification of users sharing the same behaviour irrespective of the chosen classifiers or performance metrics.