ACBM: attention-based CNN and Bi-LSTM model for continuous identity authentication

Abstract

With the evolution of network attack methods, implicit continuous identity authentication technology has attracted more and more attention. Among them, keystroke dynamics is widely used because it does not need the assistance of devices other than keyboards. In this paper, we propose a keystroke dynamic identity authentication model based on deep learning. This model combines convolutional neural network (CNN), bi-directional Long Short-Term Memory (BI-LSTM), and the attention mechanism. Unlike most existing models that only use keystroke time as the feature vector, this model uses keystroke content and keystroke time as the feature vector. First, CNN is used to process feature vectors. Then the normalized vector is input into the bi-LSTM network for training. The model in this paper is tested using Buffalo open data set. The results show that FRR (False Reject Rate), FAR (False Accept Rate), and EER(Equal Error Rate) are 3.09%, 3.03%, and 4.23%, respectively. The validity and accuracy of the model in continuous identity authentication are proved.