SynLAD-HE

# SynLAD-HE: Synthetic Learning Analytics Dataset for Higher Education **Creator**: Dr. Sanjay Agal **Affiliation**: Department of Artificial Intelligence and Data Science, Parul University, India **GitHub Repository**: https://github.com/drsanjayagal/SynLAD-HE **Publication date**: 2026-04-12 **DOI**: 10.5281/zenodo.19529000 **Version**: 1.0 --- ## Description SynLAD-HE is a large-scale, multi-table synthetic dataset designed to support advanced research in learning analytics, student success modeling, and educational data mining. The dataset comprises **over 100 million records** distributed across eight interrelated tables, simulating real-world academic, behavioral, and institutional data in higher education systems. The schema is inspired by operational patterns observed in a private university in Western India and is generalized to ensure applicability across global higher education contexts. --- ## Dataset Composition - **Students (≈5M records)** Demographics, socio-economic status, admission type - **Courses (≈50K records)** Course metadata including credits, difficulty level, and discipline - **Enrollments (≈30M records)** Student-course registrations across semesters - **Assessments (≈40M records)** Internal and external marks, assignment scores, computed grades - **Attendance (≈15M records)** Attendance percentage per student per course - **LMS Logs (≈8M records)** Learning management system interactions, time spent, resource access, login frequency - **Academic Performance (≈2M records)** GPA, CGPA, and backlog indicators - **Dropout Risk (≈500K records)** Predicted dropout probability and categorical risk levels --- ## Data Generation Methodology The dataset is generated using a hybrid simulation framework that combines: - Statistical distributions: - Gaussian (marks, GPA) - Poisson (interaction frequency) - Categorical (demographics and academic attributes) - Correlation modeling: - Higher attendance → improved academic performance - Increased LMS engagement → higher grades - Lower socio-economic status → elevated dropout risk - Controlled stochastic noise for realism - Referential integrity across all tables using primary and foreign keys - Fixed random seed (**42**) to ensure full reproducibility --- ## Use Cases - Benchmarking machine learning and deep learning models for student success prediction - Evaluating synthetic data generation techniques - Testing large-scale learning analytics pipelines - Developing early warning systems for dropout prediction - Conducting privacy-preserving research without real student data --- ## Technical Details - **Format**: CSV (8 relational tables), optional Parquet conversion - **Total Records**: >100 million - **Approx. Size (compressed CSV)**: ~15 GB - **Programming Environment**: Python 3.8+ - **Libraries Used**: pandas, numpy, faker, tqdm - **Reproducibility**: Seed-controlled data generation - **License**: MIT License --- ## File Structure - `students.csv` - `courses.csv` - `enrollments.csv` - `assessments.csv` - `attendance.csv` - `lms_logs.csv` - `academic_performance.csv` - `dropout_risk.csv` - `generate_synlad_he.py` (dataset generation script) - `README.md` (documentation and usage instructions) --- ## Citation If you use this dataset, please cite: ```bibtex@dataset{agal_synlad_he_2026, author = {Dr. Sanjay Agal}, title = {SynLAD-HE: A Benchmark Synthetic Learning Analytics Dataset for Modeling Student Success in Higher Education}, year = {2026}, publisher = {Zenodo}, doi = {10.5281/zenodo.19529000}, url = {https://doi.org/10.5281/zenodo.19529000}}