Classification and authentication of one-dimensional behavioral biometrics
John Vincent Monaco · 2014
For some behavioral biometrics, only the timestamps of a recurring event may be available. This is the case for the recently proposed random time interval (RTI) biometric in which a user repeatedly presses a single button. A dynamical systems approach is taken to deal with biometrics which are inherently one-dimensional. The methodology uses the minimum description length principle to find the optimal time delay embedding for a time series and an optimization to the multivariate Wald-Wolfowitz test for efficiently comparing time series of different lengths. Promising classification and authentication results are achieved on several experimental datasets, utilizing event timestamps only. Classification accuracy ranged from 16.2% to 44.1% and authentication EER from 32.8% to 12.7%. The proposed methodology was also used to achieve first place in the 2014 EMVIC, with 39.6% classification accuracy. All code is made available for experiment reproducibility.