Pro-Authentication Anomaly Detection Solution Based on Raw Key and Mouse Behavior Data
Liang Yin, Zhiyan Ning, Yuanzhe Lv, Xu An Wang, Xingwen Zhao · 2023
As the first layer of security for computer systems, authentication technology plays a pivotal role in protecting users’ personal information. However, the behavior anomaly of the login user after authentication should be detected because the cryptographic credentials may be leaked. Currently, behavior anomaly detection without the user's awareness is made possible by detection methods based on keystroke and mouse behaviors, which is a hot research area right now. However, the data sets of existing schemes often impose different degrees of limitations, and the identification capability needs to be improved. In this paper, we describe a scheme based on the fusion of keystroke and mouse behavior features to detect anomaly of login user, and collect data sets in the user's natural working environment without any restrictions. Inspired by image recognition, we first map the raw data to grey-scale images and then use convolutional neural networks to extract deep features from the images. Detection results of 0.990 accuracy were obtained when the input image size was 64 × 64 in size. The paper also investigates the effect of adding window features to keystroke behaviors on the detection performance.