Session-Based User Authentication via Mouse Dynamics
Metehan Yıldırım, Emin Anarım · 2019
Applying mouse dynamics to address security issues like security breaches and malicious attacks is an emerging method. The Balabit dataset is one of very few datasets shared publicly and the presence of very short test sessions makes this dataset difficult. This article presents an authentication algorithm for imposer detection using this dataset via mouse dynamics. The raw data were divided into mouse movements within each session, whereas these mouse movements were classified according to their size. Separate signatures were created for each user by calculating the individual attributes extracted from short and long size mouse movements. Then a two-class classification is employed to authentication. Session-based authentication performance reached 0.93 AUC and 0.13 EER. This study shows that session-based authentication can be performed using only mouse movement curves. Unlike other studies, the effect of a curve-length-based separation on performance is stated.