Federated Learning for Data Security and Trust in University Sports Systems

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

University sports information systems continuously aggregate data from fitness tests, courses, venue use, check-ins, and wearable devices—multi-source, heterogeneous, distributed, and privacy-sensitive data whose centralized modeling exposes students’ sports and health information, complicates cross-node collaboration, and limits scalability. We therefore build a secure, trustworthy federated learning mechanism for university sports information systems with a six-layer framework spanning data acquisition through application services. After local training it applies gradient pruning, differential-privacy perturbation, and secure aggregation, while node trustworthiness—from contribution, update stability, anomaly risk, and historical reputation—drives trust-weighted aggregation, coupling privacy protection with dynamic credibility control rather than treating them as independent modules. Under non-IID distributions, simulations show the method achieves accuracy, recall, F1, and AUC of 0.887, 0.873, 0.879, and 0.914, exceeding FedAvg’s F1 by about 5.8 points and DP-FedAvg’s by about 6.7; with 30% malicious nodes it still holds F1 = 0.831 versus 0.702 for FedAvg. The mechanism thus improves performance, privacy, and anomalous-node suppression without raw sports data leaving local nodes