Design and Assessment of Federated Digital Learning Policy Models for Sustainability-Oriented Higher Education

Abstract

The explosive growth of digital learning ecosystems within higher education has increased the requirement for policy mechanisms that can solve issues related to data privacy, institutional autonomy, inclusive participation, computational efficiency, and sustainable learning outcomes. Centralized analytics are highly dangerous from the perspective of data sovereignty, while existing federated learning algorithms give priority to performance and the number of samples without considering universities with limited resources and neglecting institutional policies concerning equity and carbon footprints. To fill this gap, this research proposes a novel Federated Sustainability Policy Evaluation and Orchestration (FedSustain-PEO) approach. FedSustain-PEO consists of a Machine Executable Policy Translation Engine that turns qualitative institutional policies into computable constraints, a Dynamic Policy Constraint Graph for representing policy conflicts, and a Policy Adaptive Federated Controller for regulating the behavior of the training process. Moreover, the Policy Aware Sustainability Aggregation technique allows balancing the multi-objective optimization with respect to five key criteria, including Policy Compliance, Sustainability Learning Outcomes, Inclusion, Privacy Risk, and Computational Costs. As evaluated on a heterogeneous testing ground of 12 institutional nodes, FedSustain-PEO showcased outstanding results relative to existing standards and baselines. It provided an accuracy of 94.8%, a precision of 94.1%, and an F1 score of 93.8%. As for its performance in governance and equity, it provided a Policy Execution Compliance Index of 0.947, a Federated Institutional Equity Score of 0.918, and cut policy violation rates to 4.8%. Further, it provided a Sustainability Learning Gain of 0.842, decreased overall energy usage by 26.5% (to 41.6 kWh), and was capable of maintaining a high participation rate of 78.6% among resource-constrained institutional nodes. In summary, FedSustain-PEO provides a scalable and privacy-protected basis for distributed university learning. By turning the policy evaluation process into a multi-objective continuous orchestration procedure, it is capable of balancing prediction utility with data privacy, equity, and sustainability.