Scholarlink

Predicting student dropout is essential for timely intervention and improving learning outcomes. However, privacy concerns limit the ability to centralize student data for deep analyses. Federated Learning (FL) provides a privacy-preserving paradigm for collaboratively training AI models. In an FL scenario, data owners learn a shared model by aggregating locally computed partial models without sharing their sensitive raw data with other parties. In this work, we compare two FL methods - Federated Averaging (FedAvg) and Federated Proximal (FedProx) - for dropout prediction in educational settings. We also integrate post-hoc explainable AI (XAI) methods, including LIME, SHAP, and integrated gradient, to improve model transparency and provide actionable insights into key dropout factors. Extensive experiments on three datasets demonstrate that FedProx slightly outperforms FedAvg. Furthermore, the results show that both FedProx and FedAvg achieve performance comparable to centralized learning in terms of F1 score and accuracy while preserving privacy. This research contributes to the development of explainable and privacy-preserving dropout prediction models, supporting informed decision-making in education. The code and all experimental results are publicly available on GitHub ( https://github.com/lamsiyah/Federated-Learning-Student-Dropout ).

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Privacy-Preserving Federated Learning for Student Dropout Prediction: Enhancing Model Transparency with Explainable AI

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Salima Lamsiyah,
  • Aria Nourbakhsh,
  • Samir El-Amrany,
  • Christoph Schommer

摘要

Predicting student dropout is essential for timely intervention and improving learning outcomes. However, privacy concerns limit the ability to centralize student data for deep analyses. Federated Learning (FL) provides a privacy-preserving paradigm for collaboratively training AI models. In an FL scenario, data owners learn a shared model by aggregating locally computed partial models without sharing their sensitive raw data with other parties. In this work, we compare two FL methods - Federated Averaging (FedAvg) and Federated Proximal (FedProx) - for dropout prediction in educational settings. We also integrate post-hoc explainable AI (XAI) methods, including LIME, SHAP, and integrated gradient, to improve model transparency and provide actionable insights into key dropout factors. Extensive experiments on three datasets demonstrate that FedProx slightly outperforms FedAvg. Furthermore, the results show that both FedProx and FedAvg achieve performance comparable to centralized learning in terms of F1 score and accuracy while preserving privacy. This research contributes to the development of explainable and privacy-preserving dropout prediction models, supporting informed decision-making in education. The code and all experimental results are publicly available on GitHub ( https://github.com/lamsiyah/Federated-Learning-Student-Dropout ).