A Multi-Campus Academic Performance Prediction Algorithm Based on Federated Machine Learning and Neural Network

In a multi-campus teaching model, accurate prediction of student academic performance facilitates rational allocation of teaching resources and personalized instruction. However, due to data privacy and security concerns, direct data sharing across campuses is infeasible. To address this, we propose a federated learning and neural network-based algorithm for multi-campus educational performance prediction. Each campus stores data locally and builds models independently, transmitting only model update parameters instead of raw data. Combined with encryption and distributed optimization techniques, the approach enables localized processing while a neural network learns nonlinear mappings between performance, course difficulty, learning behaviors, and historical grades. Experimental comparisons on multiple datasets demonstrate that the method significantly outperforms traditional centralized machine learning in accuracy while preserving data privacy. The algorithm also allows flexible adjustments, ensuring compatibility and scalability across campuses, thus offering a feasible solution for multi-campus educational informatization.