Exploration of Feature Representations for Predicting Learning and Retention Outcomes in a VR Training Scenario

Abstract

Training and education of real-world tasks in Virtual Reality (VR) has seen growing use in industry. The motion-tracking data that is intrinsic to immersive VR applications is rich and can be used to improve learning beyond standard training interfaces. In this paper, we present machine learning (ML) classifiers that predict outcomes from a VR training application. Our approach makes use of the data from the tracked head-mounted display (HMD) and handheld controllers during VR training to predict whether a user will exhibit high or low knowledge acquisition, knowledge retention, and performance retention. We evaluated six different sets of input features and found varying degrees of accuracy depending on the predicted outcome. By visualizing the tracking data, we determined that users with higher acquisition and retention outcomes made movements with more certainty and with greater velocities than users with lower outcomes. Our results demonstrate that it is feasible to develop VR training applications that dynamically adapt to a user by using commonly available tracking data to predict learning and retention outcomes.