Federated learning-enabled personalised delivery and student privacy protection in universities

This study presents a federated learning framework enhanced with entropy-adaptive differential privacy, blockchain consensus, and knowledge distillation to safeguard student data while improving personalised education.Traditional federated learning preserves privacy by training collaboratively without sharing raw data, yet faces challenges of heterogeneity, efficiency, and resilience against malicious clients.Existing solutions like homomorphic encryption and secure multiparty computation often incur high computational costs and limited adaptability.To address these limitations, the proposed framework employs blockchain-based role incentives to ensure fairness and verifiability, while entropy-adaptive differential privacy dynamically balances privacy and utility.Knowledge distillation further improves robustness and mitigates non-IID data distribution issues.Experiments on a Python programming course dataset with 2,452 students demonstrate superior accuracy, fairness, and resilience compared to conventional FedAvg.The method achieves up to 97% prediction accuracy with enhanced stability under adversarial conditions, offering a scalable and secure solution for personalised, privacy-preserving education.