A Privacy-Preserving Federated Digital Twin Educational Metaverse Framework for Secure Personalized Learning

Abstract

While the use of educational metaverse platforms enables immersive and personalized learning, it also generally relies on centralized processing of learner data, which raises privacy, security, and governance concerns. This paper presents FPDT-EMF, a privacy-preserving federated digital-twin educational metaverse framework, combining federated learning, digital-twin learner modelling, reinforcement learning (RL) personalization, secure aggregation, differential privacy (DP), and explainable artificial intelligence (XAI). The framework enables educational institutions to train learning models while keeping the raw learner data locally. The experiments were carried out with a sample of 800 students distributed among 20 institutional clients and over 15,000 learning sessions. The accuracy, precision, recall, and F1 score of FPDT-EMF were 98%, 97%, 96%, and 96%, respectively. It also achieved a high privacy rate of 99%, security rate of 98%, learner trust rate of 97%, personalization effectiveness rate of 98%, and scalability rate of 97%. Membership-inference evaluation has an AUC of 0.52, which means there is a moderate risk of privacy leakage. Ablation and statistical analyses confirmed the contribution and significance of the proposed modules. The results indicate that FPDT-EMF can be the secure, scalable, transparent, and personalized learning framework for next-generation learning metaverse environments.