Cross-Domain Gender Identification Using VR Tracking Data
Qidi J. Wang, Alec G. Moore, Nayan N. Chawla, Ryan P. McMahan · 2024
Recently, much work has been done to research personal identifiability of extended reality (XR) users. Many of these prior studies are task-specific and involve identifying users completing a specific XR task. On the other hand, some studies have been domainspecific and focus on identifying users completing different XR tasks from the same domain, such as watching 360° videos or assembling structures. In this paper, we present one of the few studies to investigate cross-domain identification (i.e., identifying users completing XR tasks from different domains). To facilitate our investigation, we used open-source datasets from two different virtual reality (VR) studies-one from an assembly domain and one from a gaming domain-to investigate the feasibility of cross-domain gender identification, as personal identification is not possible between these datasets. The results of our machine learning experiments clearly demonstrate that cross-domain gender identification is more difficult than domain-specific gender identification. Furthermore, our results indicate that head position is important for gender identification and demonstrate that the k-nearest neighbors (kNN) algorithm is not suitable for cross-domain gender identification, which future researchers should be aware of.