Research On Continuous Authentication Scheme Based On Raw Data Of Keystroke And Mouse Behavior

The rapid development of Internet technology is driving the progress of human society,but the number of network security incidents is also increasing day by day.As the first layer of security for computer systems,authentication technology plays a pivotal role in the protection of users’ personal information.Traditional authentication technologies are onetime authentication,i.e.,the user’s identity is authenticated at a certain moment when the user logs in,and the user’s identity cannot be continuously authenticated.Continuous authentication refers to the continuous authentication of the user’s identity throughout the session cycle of the user’s use of the computer.Biometric-based authentication method uses unique biometric features to identify users,which is a stable and irreproducible method and is a hot topic of research today.Biometric features can be divided into physiological features and behavioral features,such as appearance,palm print,etc.,and behavioral features such as the user’s speaking voice,gait,keystroke and mouse behavior,etc.The authentication method based on user’s physiological characteristics often requires additional data collection devices to collect user’s physiological characteristics data,which cannot achieve continuous authentication,while the authentication method based on user’s keyboard and mouse behavior can continuously authenticate user’s identity without additional hardware devices.Therefore,this paper conducts an exploratory study on the continuous authentication scheme based on user’s keystroke and mouse behavior.The main work accomplished in this paper is as follows:(1)A dataset based on raw data of user’s keystroke behavior was collected and produced,and a data collection program was written to collect raw data of user’s keystroke behavior without any restrictions on user’s behavior,and a dataset of images of user’s keystroke behavior was produced by mapping raw data of user’s keystroke behavior to grayscale images in order to authenticate user’s identity by extracting deep features from the images using convolutional neural networks.(2)A mouse dynamics-based authentication scheme is designed.In this paper,we first train and test the VGG16 network model on the publicly available Balabit mouse dataset and obtain an accuracy rate of 0.996,a False Accept Rate(FAR)of 0.33% and a False Reject Rate(FRR)of 0.34% in about 36.206 seconds.Secondly,based on migration learning,the network parameters learned by the Inception V3 network in the large image dataset were applied to the Balabit dataset for training,and the authentication time was also 36.206 s,but the training model time was greatly reduced,and the final accuracy of 0.993 was obtained,and the FAR and FRR were 0.942% and 0.3947%,respectively,for the two schemes are improved compared with the same type of schemes by other scholars;(3)A keystroke behavior data-based authentication scheme is designed.Considering that the single-factor mouse-based authentication may give intruders an opportunity to take advantage of it,this paper studies a scheme based on the fusion of keystroke behavior features to authenticate users,and based on the keystroke behavior dataset produced in this paper and the improved ResNet18 network,we obtain an accuracy of 0.990% when the input image size is 64x64 size.Based on the keystroke behavior dataset and the improved ResNet18 network,the authentication result is 0.990,0.2% and 0.3% of FAR and FAR respectively,and the authentication time is about 24.046 s.Finally,we also verify the influence of the window features corresponding to the keystroke behavior on the performance of the authentication model,and find that the influence of the window features on the performance of the model becomes smaller and smaller when the input image size is larger than 64,which can provide a reference for later researchers.This empirical experience can provide a reference for later researchers in program design.