A Real-Time Digital Twin Framework for Personalized Learning Optimization: Integrating Multimodal Behavioral Data, Cognitive State Estimation, and Explainable AI

Personalized learning has been widely promoted as a pathway to improve educational outcomes, yet most existing systems remain limited by static learner models, delayed feedback, and weak interpretability. This study introduces a real-time Digital Twin-based educational framework that continuously mirrors the cognitive, behavioral, and performance states of individual learners to enable dynamic and adaptive learning optimization. The proposed system integrates multimodal data streams, including interaction logs, temporal engagement patterns, and performance trajectories, to construct a continuously updated learner-specific digital replica. A hybrid modeling architecture is developed, combining temporal deep learning models with probabilistic state estimation to infer latent cognitive states such as cognitive load, attention fluctuation, and learning stability. To address the critical issue of transparency in AI-driven educational decisions, the framework incorporates explainable artificial intelligence techniques, allowing both instructors and learners to interpret system recommendations and predicted outcomes. The framework is evaluated using a dataset of ۴۲۸ undergraduate students over a ۱۲-week semester in a blended learning environment. Results demonstrate that the Digital Twin-based system improves prediction accuracy of academic performance by ۲۳.۷% compared to baseline machine learning models, while reducing late-stage learning failure risk by ۳۱.۴% through early intervention mechanisms. In addition, adaptive content delivery guided by the digital twin model leads to a statistically significant improvement in learning gain (p < ۰.۰۱), with an average increase of ۱۸.۶% in normalized learning scores. The findings highlight the potential of Digital Twin technology as a transformative paradigm in educational psychology, enabling a shift from reactive to predictive and proactive learning systems. The study also discusses implementation challenges, including data privacy, model generalizability, and computational scalability, providing a roadmap for future research in intelligent and human-centered educational systems.

کلیدواژه ها:

Digital Twin in Education ، Personalized Learning ، Educational Data Mining ، Cognitive State Estimation ، Explainable Artificial Intelligence ، Adaptive Learning Systems ، Multimodal Learning Analytics ، Predictive Learning Models