Mouse dynamics based user recognition using deep learning

Abstract

Behavioural biometrics provides an extra layer of security for user authentication mechanisms. Among behavioural biometrics, mouse dynamics provides a non-intrusive layer of security. In this paper we propose a novel convolutional neural network for extracting the features from the time series of users’ mouse movements. The effect of two preprocessing methods on the performance of the proposed architecture were evaluated. Different training types of the model, namely transfer learning and training from scratch, were investigated. Results for both authentication and identification systems are reported. The Balabit public data set was used for performance evaluation, however for transfer learning we used the DFL data set. Comprehensive experimental evaluations suggest that our model performed better than other deep learning models. In addition, transfer learning contributed to the better performance of both identification and authentication systems.

Machine and Deep Learning Applications to Mouse Dynamics for Continuous User Authentication

“…With regards to CNNs specifically, liberties have been taken with the representation of data for mouse dynamics, as an optimal representation has not yet been found due to the novelty of the field. Some suggestions include using speed rather than absolute coordinates to ensure a translationally invariant model [25] or graphing mouse movements on a 2-dimensional plane and using a traditional 2D-CNN [24]. Many of these methods are compatible with our data, as Figure 2 outlines.…”

“…However, instead of using a timestep of 3, we simply pass a single mouse action as input to the model. Since The architecture of the model follows closely to [25], however is shallower in width. Model sizes of varying capacities were tested, however yielded negligible improvements in overall accuracy.…”

“…By focusing on and extracting the spatial features of mouse dynamics data, the image processing capabilities of CNNs can be similarly applied to user authentication. However, the strengths of convolutions in CNNs extend beyond spatial features, as the temporal approach of modeling data as a time-series signal and using a 1D-CNN to convolve along the singular time dimension of the signal has been proposed as well [25]. Further, [25] uses X and Y velocities to create signals from their mouse events rather than absolute coordinates to produce a translation invariant model.…”

“…However, the strengths of convolutions in CNNs extend beyond spatial features, as the temporal approach of modeling data as a time-series signal and using a 1D-CNN to convolve along the singular time dimension of the signal has been proposed as well [25]. Further, [25] uses X and Y velocities to create signals from their mouse events rather than absolute coordinates to produce a translation invariant model. Consequently, recurrent neural networks -especially their long short-term memory variants (LSTM-RNN) -have been proposed for use due to their known strong performances with time-related data [26,27].…”

SapiAgent: A Bot Based on Deep Learning to Generate Human-Like Mouse Trajectories

“…In recent works using deep neural networks, mouse trajectories were divided into blocks of fixed length. Promising results of user authentication were presented by Chong et al [16] and Antal and Fejer [17]. In contrast, Tan et al [18] presented different adversarial attack strategies against behavioral mouse dynamics.…”