RUMBA-Mouse: Rapid User Mouse-Behavior Authentication Using a CNN-RNN Approach

Mouse behavior analysis has become an increasingly attractive area to biometric researchers in recent years. Many mouse behavior based user authentication schemes have been proposed in the past decades. However, most of them rely on statistical analysis or heuristic feature extraction of the mouse behavior. In this paper, we present a CNN-RNN combined neural network model for mouse behavior based user authentication, which takes raw sequential mouse data as input rather than relies on heuristic feature extraction. Additionally, we integrate the model with a practical framework of static user authentication and evaluate it on a real dataset. The results show our approach yields a 3.16% EER and a 99.39% AUC, with a short authentication delay of 6.11 seconds on average, which demonstrates the effectiveness and practicality of applying deep learning techniques for static mouse behavior based user authentication. Furthermore, by modifying the activation maximization method, we study and visualize the features learned by different layers of our neural network.