PAITS: Detecting Masquerader via Short-Lived Interventional Mouse Dynamics

It is relatively easier for an insider attacker to steal the password of a colleague or use an unattended machine (logged in by other users) within a trusted domain to launch an attack. A simple real-time authentication by password may not work if they have the password. By comparing the stored mouse behavioral profile of the valid user, the system automatically authenticates the user. However, long verification time in existing approaches based on mouse dynamics which mostly last dozens of minutes and probably make masquerader escaped from detection mechanism. In this paper, we proposed a system called PAITS (Practical Authentication with Identity Tracing System) to do re-authentication via comparison of mouse behavior under a short-lived interventional scenario. Mouse movements under the special scenario where the cursor is a bit out of control can capture the user’s unconscious reaction, and then be used for behavioral comparison and detection of malicious masquerader. Our experiments on PAITS demonstrate best result with a FRR of 2.86% and a FAR of 3.23% under probability neural network with 71 features. That is a comparative result against the previous research results, but at the same time significantly shorten the verification time from dozens of minutes to five seconds.