EUAS-GAN: Enhancing User Authentication on Smartphones Through GAN-Based Swiping Data Augmentation

Generative Adversarial Networks (GANs) have demonstrated exceptional potential in synthetic media generation but remain underutilized in the domain of behavioural biometrics. This study presents EUAS-GAN, a framework that leverages GANs to generate high-quality synthetic swiping data, aiming to enhance the performance of machine learning classifiers. Using a publicly available dataset, the scheme achieves an accuracy of 91.64% with 30 real training samples, which improved to 92.95% with additional 30 synthetic samples, using Random Forest as classifier. A proof-of-concept implementation demonstrated the system’s practicality, followed by a real-world usability evaluation using the standard System Usability Scale (SUS), yielding a score of 76.57. Smartphone-based testing further confirmed its feasibility, establishing EUAS-GAN as an accurate and user-friendly solution for behavioural biometric-based authentication.