A Cross‐Domain Evaluation Method for Multi‐Institutional Collaborative Student Practical Education Projects Based on Federated Learning

ABSTRACT

To address the challenges of data privacy protection and cross‐domain evaluation in multi‐university collaborative projects for students' practical education, a cross‐domain competence evaluation method for such projects based on federated learning is proposed. A three‐layer practical teaching system (basic knowledge layer, professional skills layer, and innovative design layer) is constructed, and a four‐dimensional competence evaluation index system (covering basic skills, professional skills, innovative ability, and collaborative ability) is established. A lightweight federated learning framework with heterogeneous device adaptability is adopted, which combines an adaptive federated Dropout mechanism to dynamically generate sub‐models. Homomorphic encryption functions are used to encrypt local competence evaluation vectors, and two‐level secure aggregation is realized through edge aggregators and a central server. A grouped aggregation verification mechanism is introduced to ensure model quality and eliminate abnormal nodes. Experiments show that this method achieves an average data encryption security rate of 99.05% in a simulated collaboration among 3 universities, with a data transmission packet loss rate always below 20%. It effectively evaluates 8 indicators (such as project completion rate, with a maximum of 95.43%, and innovation index, with a maximum of 95.12 points) across multiple scenarios, including technological innovation, social services, and cultural inheritance, and supports cross‐university competence benchmarking analysis. Ultimately, a secure, reliable, accurate, and efficient cross‐domain competence evaluation system for multi‐university collaborative projects in students' practical education is built, providing core federated learning support for educational resource sharing and cross‐university collaboration optimization.