Confidence curve for continuous authentication

Abstract

In this paper, we present an algorithm that continuously measures a user’s legitimacy by monitoring his behavior in real time using his mouse activity. The proposed algorithm can also be used to analyze a user’s log files and determine if it is the right user and if his behavior can be described as normal. We built our own database, with employee data from a computer company. We use Support Vector Machine (SVM), a supervised machine learning algorithm, to detect abnormal behavior as quickly as possible. A graph, representing the confidence level, is also drawn in real time to visualize the confidence level given to the user. We save it in case we need to check the user’s legitimacy later. Experimental results show that our system can accurately authenticate a user and quickly detect an impostor. We can either set confidence thresholds that block the computer if the confidence curve decreases below or just monitor the tendency of the curve. A significant advantage is the ease of analyzing the result and also that we have tested it on real data.