Modeling students’ self-regulated learning from virtual reality-integrated e-module interaction data

This study investigates how interaction data generated from virtual reality (VR)-integrated e-modules can be utilized to model students’ self-regulated learning (SRL) through an explainable machine learning approach. Grounded in the multidimensional framework of SRL, the study transforms behavioral trace data into theoretically informed features representing planning, monitoring, and reflection processes. A quantitative cross-sectional design was employed involving 138 elementary students who participated in four structured VR-supported learning sessions in Jakarta, Indonesia. Behavioral interaction logs were collected alongside self-reported SRL questionnaire data. Theory-guided feature engineering was applied to align interaction behaviors with SRL dimensions before developing predictive models using Random Forest Regression and Support Vector Regression, with a mean predictor serving as the baseline model. Model performance was evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), coefficient of determination (R 2 ), and five-fold cross-validation. Both machine learning models substantially outperformed the baseline predictor, with Random Forest achieving the best predictive performance (MAE = 0.67, RMSE = 0.89, R 2 = 0.85). Feature importance analysis revealed that content revisits, post-task review duration, and feedback access frequency were the strongest predictors of SRL, indicating that monitoring- and reflection-related behaviors contributed more substantially than planning-related behaviors. Although the predictive models demonstrated stable performance across cross-validation, the highly imbalanced SRL distribution suggests that the results should be interpreted cautiously. The findings demonstrate that interaction data can provide meaningful yet partial representations of students’ self-regulated learning and highlight the importance of combining theory-guided feature engineering with explainable machine learning to support educational interpretation rather than prediction alone.