What question did this study set out to answer?

The research aims to develop a synthetic dataset that balances the need for learning analytics with privacy concerns in higher education.

March 26, 2026Open Access

A privacy preserving synthetic learner dataset for learning analytics in technology enhanced higher education

Key Points

The research aims to develop a synthetic dataset that balances the need for learning analytics with privacy concerns in higher education.
Created SynEdu-HEDL using a five-phase methodological framework.
Utilized conditional tabular generative adversarial networks and temporal sequence generators.
Incorporated differential privacy mechanisms to protect student data.
Evaluated dataset across privacy protection, statistical fidelity, and analytical utility.
Privacy guarantees demonstrated with membership inference AUC-ROC of 0.512.
Preserved essential statistical properties with an average Wasserstein distance of 0.043.
Models achieved performance within 1-5% of those trained on real data.
Transfer learning showed a 24.4% performance improvement using only 10% real data.

Abstract

The proliferation of technology-enhanced learning in higher education generates unprecedented student data, creating opportunities for learning analytics while raising critical privacy concerns. Current data sharing practices are severely constrained by privacy regulations and ethical considerations, impeding collaborative research and methodological advancement. This study addresses this fundamental tension by introducing SynEdu-HEDL, a comprehensive privacy-preserving synthetic dataset specifically designed for learning analytics in higher education.Developed through an innovative five-phase methodological framework integrating conditional tabular generative adversarial networks, temporal sequence generators, and differential privacy mechanisms, the dataset captures authentic educational patterns while ensuring robust privacy protection. SynEdu-HEDL comprises 20,000 synthetic student records encompassing 85 features across demographic characteristics, temporal learning interactions, engagement patterns, and academic performance metrics.A comprehensive three-dimensional validation framework evaluated SynEdu-HEDL across privacy protection, statistical fidelity, and analytical utility. Results demonstrate that the dataset provides strong privacy guarantees (membership inference AUC-ROC=0.512, statistically indistinguishable from random guessing), preserves essential statistical properties (average Wasserstein distance=0.043, correlation matrix similarity=94.1%), and supports diverse learning analytics tasks with models achieving performance within 1-5% of those trained on original data. Notably, transfer learning experiments show 24.4% performance improvement with only 10% real data, demonstrating practical value for resource-constrained settings.SynEdu-HEDL advances educational data science by providing a practical solution to data sharing barriers, supporting reproducible research, and establishing methodological standards for synthetic educational data validation. SynEdu-HEDL is openly available at https://github.com/drsanjayagal/SynEdu-HEDL and has been archived in Zenodo with the permanent DOI https://doi.org/10.5281/zenodo.18884938. The repository includes comprehensive documentation, fostering community engagement and accelerating progress in privacy-preserving learning analytics.

Bookmark

View Full Paper