Performance Prediction in Virtual Reality Co-Creation Learning Environments: Two-Stage vs. One-Stage Feature Selection for Student Engagement

• In Event: Teaching and Learning With Extended Reality

Abstract

This study examined the impact of feature selection on learning outcome prediction in virtual reality (VR) co-creation learning environments. 83 students attended 18-week VR co-creation courses at a northern Taiwan public high school with a blended online-merge-offline model. Objective multimodal data, including log data, quiz scores, electroencephalography (EEG), and demographics, were used for one-stage and two-stage Random Forest (RF) models predicting pass/fail learning outcomes. The two-stage RF model, with feature selection, demonstrated superior accuracy. Additionally, the RF model achieved high accuracy in predicting outcomes using EEG data, highlighting the potential of integrating physiological responses as multimodal data in learning prediction. However, the study also revealed limitations in solely relying on traditional quiz scores to assess cognitive engagement.

Authors

• Hsin-Yun Wang, National Yang Ming Chiao Tung University

• Jerry Chih-Yuan Sun, National Yang Ming Chiao Tung University