Collaborative Analysis of Multi-School Midwifery Teaching Data Driven by Federated Learning

Midwifery teaching data from multiple universities are often dispersed across institutions, forming data silos that hinder cross-university teaching analysis and optimization while increasing privacy risks under centralized processing. To address this problem, this paper constructs a cross-university collaborative analysis framework based on federated learning. The framework first uniformly encodes and standardizes midwifery teaching data features across universities to ensure cross-institutional data comparability. Each university then locally trains a deep neural network to extract teaching features, uploading only model parameters to a central server for secure aggregation. This process constructs a cross-university collaborative model while protecting student privacy. The model dynamically adjusts training weights to accommodate data heterogeneity and maintains continuous learning through incremental updates. The aggregated model generates teaching pattern analysis and optimization recommendations, providing a scientific basis for curriculum design and teaching improvement. Experimental results show that the proposed method achieves accuracy of 0.93 and F1-score of 0.92, significantly outperforming comparison algorithms. After teaching optimization, theoretical scores improve by an average of 8.2% and clinical pass rates by 10.0%. Accuracy fluctuations remain within ±0.5% to ±1.5%, validating the stability and reliability of the framework.