Gmail Multimodal Dataset (GMD)

Increased usage of technology and the Internet in our everyday lives has increased the vulnerability of our personal data and computer-based systems, motivating the need for new authentication systems that are resistant to impostor attack. Behavioral biometrics authentication utilizing unique behavioral features has emerged as a promising additional factor of authentication. However, progress in this area depends on the availability of representative public datasets. Existing datasets are often limited by highly constrained or artificial collection settings, small participant samples, or limited modality coverage, making it challenging to evaluate realistic, multimodal systems. To address these limitations, we collected a novel 56-user multimodal dataset that consists of mouse, keystroke, widget interaction, and scroll data, collected concurrently as users completed common Gmail tasks.