One v/s All SVM Implementation for Keystroke based Authentication System

Abstract

Authentication systems have laid the foundation for validating and securing user's identity. Due to increasing vulnerabilities, the traditional methods like passwords, PINs, tokens etc. cannot keep up with the challenges. The behavioural biometrics like Keystroke dynamics is used to authenticate a legitimate user through typing patterns. In this paper, we propose a machine learning approach to develop an authentication system that provides more robust user identity information using keystroke dynamics biometrics. One-v/s-all classification is applied to 51 users of CMU dataset. For optimization, the SVM classifier is used along with Grid Search for parameter tuning. Grid Search satisfies the goal of choosing the best kernel parameter pair. Experimental evaluation in a heterogeneous environment yields a false acceptance rate (FAR) of 0.2% and a false rejection rate (FRR) of 10.24%. The graphical representation of the result is expressed through ROC (Receiver Operating Characteristic) curve which shows the increased reliability of the system for authentication purpose.