HTM-MDICE: a transformer-based model for predicting student engagement and ideological understanding in ethical education

Tailoring individualized learning experiences depends on predicting student involvement and ideological awareness in ethical education, which is still difficult given educational data sets' complexity and class imbalance. HTM-MDICE, a new Transformer-based model meant to solve these issues by using hierarchical temporal modeling on a multi-modal ethical dataset of 68,200 scenarios, 1,000,000 numerical data points, and 500,000 behavioral logs, is presented in this paper. HTM-MDICE, utilizing a thorough evaluation framework, obtained a validation accuracy of 97.5%, an F1-score of 0.96, and an MAE of 0.12 with an early stopping patience of 5, therefore greatly outperforming four previous techniques-BSA-ANN, Decision Tree, BPNN, Petri Nets 10.5% in accuracy (<i>p</i> < 0.05). While preprocessing, early stopping, and the Transformer design were shown to be major factors in HTM-MDICE's performance, statistical analysis using paired <i>t</i>-tests verified the strength of its enhancements. Though it has improved, ethical issues around misclassification and data privacy call for prudent use. With future goals comprising improved interpretability, varied data integration, and longitudinal effect studies to further promote individualized education, this study adds a state-of-the-art model and assessment approach to educational predictive modeling.