Construction of a distributed collaborative framework for digital learning resources based on federated learning

Digital learning platforms generate large volumes of student interaction data, yet privacy concerns and fragmented institutional storage limit centralized analysis and personalized educational support. Federated learning (FL) offers a collaborative solution but often suffers from slow convergence and limited knowledge transfer across institutions.

The research proposes an Adaptive Optimization–Knowledge Distillation Federated Averaging (AO-KD-FA) model to enhance privacy-preserving forecast of student performance in distributed digital learning environments. The framework integrates adaptive optimization for efficient parameter tuning, knowledge distillation for secure knowledge transfer, and federated averaging for collaborative global model aggregation. Experiments were conducted using a multi-institutional student dataset with demographic, engagement, and academic performance features.

The proposed model achieved accuracy (97.30%), precision (96.80%,) recall (96.20%), and a F1-score (96%), outperforming conventional models such as DNN, Random Forest, and KNN. Cross-validation results also demonstrated stable convergence and consistent predictive performance across distributed learning environments.

These findings indicate that the AO-KD-FA framework effectively improves personalized learning prediction while maintaining strong data privacy and scalability in collaborative educational systems. The approach supports intelligent adaptive education by enabling secure knowledge sharing and efficient federated optimization across multiple institutions.