BeCAPTCHA-Mouse: Synthetic mouse trajectories and improved bot detection

We first study the suitability of behavioral biometrics to distinguish between computers and humans, commonly named as bot detection. then present BeCAPTCHA-Mouse, a detector based on: i) neuromotor model mouse dynamics obtain novel feature set for classification human samples; ii) learning framework involving real synthetically generated trajectories. propose two new trajectory synthesis methods generating realistic data: a) function-based method on heuristic functions, b) data-driven Generative Adversarial Networks (GANs) in which Generator synthesizes human-like trajectories from Gaussian noise input. Experiments are conducted testbed also introduced here available GitHub: BeCAPTCHA-Mouse Benchmark; useful research detection other mouse-based HCI applications. Our benchmark data consists 15,000 including 58 users with various levels realism. experiments show that is able detect high realism 93% accuracy average using only one trajectory. When our approach fused state-of-the-art dynamic features, increases relatively by more than 36%, proving fast, easy, reliable tool complement traditional CAPTCHA systems.