A novel approach to classify users based on keystroke behavior

Abstract

The contemporary society has been posing several security threats with evolution of modern devices. Keystroke dynamics is a behavioral biometrics that can be bolstered with the existing traditional password typing to validate the identity of the user. In the previous research a three dimensional model was constructed based on dwell time, pressure and tolerance. This paper extends the previous research focusing to improve the classification accuracy. It makes use of an unsupervised technique called self-organizing maps to cluster the data from a three dimensional space to a two dimensional space followed by a learning vector quantization algorithm to classify the user and learn from the feedback obtained. It provides simple but efficient stochastic learning to improve classification accuracy. Results show that the proposed approach leverages existing techniques with higher precision in identifying the user.