On the Improvements of Mouse Dynamics Based Continuous User Authentication

Traditional authentication methods are vulnerable when users leave their devices unattended or if their credentials are compromised. In contrast, continuous authentication offers a perpetual strategy for user validation, ensuring that only authorized users access critical information throughout their entire usage. The problem of continuous authentication boils down to a binary classification task: determining whether the usage is legal or illegal. Deep learning presents a promising solution for this problem, although the use of convolutional neural networks (CNNs) in continuous authentication still has room for improvement. In this study, we employ residual learning to train and test a user authentication model. To further enhance the accuracy of the results, we implement a realistic augmentation method and employ a superior image mapping technique compared to existing literature. As a result, we achieve significantly more accurate results than those reported in the referenced studies. On average, our tests yield a False Accept Rate of 0.45 and a False Reject Rate of 0.34, which are 6.5 times better than the referenced studies. These findings demonstrate a substantial improvement in the usability and effectiveness of real-world cybersecurity applications.