Privacy-Preserving Student Behavior Modeling Based on Federated Learning

Abstract

In the context of the digitalization of higher education, the analysis of student behavioral data has become a key instrument for improving the quality of education, personalizing learning trajectories, and enabling early detection of academic risks. However, centralized collection and processing of educational data create significant privacy threats, including personal data leaks, regulatory violations, and a decline in student trust. This study proposes a student behavior prediction model based on federated learning, ensuring data privacy while maintaining high predictive accuracy. The proposed architecture integrates federated aggregation mechanisms, differential privacy, and secure gradient exchange, enabling the construction of a distributed learning system without transferring raw data to a central server. A mathematical model of the student behavioral profile is developed, incorporating academic activity, digital traces in Learning Management Systems (LMS), temporal interaction patterns, and engagement indicators. The experimental component of the study includes simulations using both synthetic and real educational datasets, applying neural networks and gradient boosting within a federated learning environment. The results demonstrate that the federated model achieves comparable predictive accuracy to centralized training (with a difference of less than 2.3%), while significantly reducing the risk of personal data disclosure. Additional robustness analysis against gradient inversion attacks confirms an increased level of information protection. The findings support the feasibility of deploying federated learning in educational analytics systems and open new opportunities for developing ethically sustainable intelligent educational platforms. The practical value of the proposed approach lies in its ability to be integrated into existing Learning Management Systems (LMS) and national educational platforms. Under increasingly strict regulatory requirements regarding personal data processing, universities are compelled to adopt architectures that minimize the transmission of sensitive information. Federated learning enables inter-university analytics without the need for centralized data storage, significantly reducing legal and reputational risks.