POHMM/SVM: A Hybrid Approach for Keystroke Biometric User Authentication

Abstract

There are mainly two kinds of statistical models are found in keystroke biometrics namely discriminative model and generative model. The support vector machine (SVM) is a popular discriminative models used in keystroke biometric systems for the last decade due to a higher accuracy rate for large datasets. On the other hand., the hidden Markov model (HMM)., a generative model., has proven to be a useful and efficient tool., especially in speech recognition. However., its performance is poor in keystroke biometrics compared to other models.. An extension of HMM - partially observable hidden Markov model (POHMM) has shown better performance in handling missing or infrequent data. In an attempt to reach efficiency., this study proposes a hybrid POHMM/SVM approach for user authentication taking advantage of both generative and discriminative models. POHMM has been used as the features extractor., and the one-class support vector machine as the anomaly detector. The proposed POHMM/SVM model has achieved 0.086 of average equal-error rate (EER) on CMU keystroke benchmark dataset across all subjects and the standard deviation is 0.063. Using the same evaluation procedure described in the supplemental paper for CMU benchmark keystroke dataset., the proposed model has shown substantial decrease in the EER over other published methods.