VPSD: A Multimodal Virtual Presentation Dataset for Student Speaking Proficiency Classification in Online Learning Environments

The Virtual Presentation Speaking Dataset (VPSD) is a large-scale multimodal educational dataset designed for speaking proficiency classification in virtual presentation environments. The dataset contains 485,716 session-level presentation records with four speaking proficiency categories: Beginner, Intermediate, Advanced, and Expert. VPSD includes heterogeneous multimodal educational features derived from speech, transcript, visual-behavioral, and presentation-interaction modalities. The dataset contains communication-related descriptors such as pronunciation accuracy, semantic coherence, fluency behavior, speaking rate, pause characteristics, gesture consistency, eye-contact behavior, posture stability, facial confidence, audience interaction quality, slide-transition smoothness, and presentation engagement indicators. The dataset consists of structured numerical educational representations suitable for multimodal learning, educational analytics, speaking proficiency classification, communication assessment, and explainable artificial intelligence research in online learning environments.