Deep learning–based behavioral biometric authentication using keystroke and touch patterns

Abstract

With the development of digitization, sensitive data protection and online transactions have become crucial. Traditional authentication methods, such as using a PIN and password on a regular basis, have to deal with credential theft, phishing, and brute force attacks. Unlike static authentication systems like knowledge-based authentication or physical biometric authentication, behavioral biometric authentication focuses on one’s unique and habituated interactions with one’s own device. This approach does not use anything that a user knows or has as a physical presence. Rather, it uses something that a user may as well be unaware of but follows on a daily basis, which is keystroke pattern and touch pattern. The technique fits well in the task of enhancing user verification without sacrificing usability. This research extends the domain of access control through the design of a flexible and scalable authentication system. For the assessment of the effectiveness of this technique, a cross-device web platform was built. By which keystroke and touch input of 30 users in four different sessions have been collected. Machine learning analysis of keystroke and touch behavior confirms the effectiveness of behavioral biometrics for identity verification. The initial findings from the model, like those of Siamese LSTM, demonstrate high accuracy (93% training accuracy and 82% test accuracy) are encouraging for keystroke. As a result, our solution makes sense as a substitute for conventional authentication methods.