SapiMouse: Mouse Dynamics-based User Authentication Using Deep Feature Learning

Abstract

The increasing interest in the analysis of mouse-based human-computer interaction may be attributed to prominent applications, such as user authentication and bot detection based on mouse dynamics. The aim of our paper is to present the SapiMouse dataset that can be used for training and evaluation of both user authentication and bot detection systems. In this paper we present the tools and protocols used for data collection, as well as the exploratory analysis of this new dataset. In addition, we also present user authentication results on this new dataset. Instead of using handcrafted features, our system learns the features directly from raw data using a convolutional neural network. The performance of our system is 0.94 AUC for 15 seconds of data.