Applying feature selection to reduce variability in keystroke dynamics data for authentication systems

Abstract

Authentication systems enable the verification of claimed identity. Password-based authentication systems are ubiquitous even though such systems are amenable to numerous attack vectors and are therefore responsible for a large number of security breaches. Biometrics has been increasingly researched and used as an alternative to password-based systems. There are a number of alternative biometric characteristics that can be used for authentication purposes, each with different positive and negative implementation factors. Achieving a successful authentication performance requires effective data processing. This study investigated the use of keystroke dynamics for authentication purposes. A feature selection process, based on normality statistics, was applied to reduce the variability associated with keystroke dynamics raw data. Artificial Neural Networks were used for classification, and results were calculated as the false acceptance rate (FAR) and the false rejection rate (FRR). Experimental results returned an average FAR of 0.02766 and an average FRR of 0.0862, which were at least comparable with other research efforts in this field.