Data for: Prediction Without Intervention? A Systematic Review and Empirical Benchmark of Learning-Analytics Early-Warning Systems for Student Retention in Higher Education

This dataset contains the underlying data for a study that pairs a PRISMA 2020 systematic review with an original, reproducible empirical benchmark on learning-analytics early-warning systems for student retention and dropout in higher education.

The review component comprises the bibliographic search log, the corpus of records retrieved from the OpenAlex database (2015-2026), the record-level screening decisions (include/exclude, analytic-and-decision category, study design, outcome-reporting flag, relevance), the list of 176 included studies, the prioritised synthesis set, the subset of studies reporting a retention outcome, the PRISMA flow counts, and the computed descriptive bibliometric results.

The benchmark component contains the public student dataset used for the empirical analysis (Realinho et al., 2022; 4,424 students, 36 features) together with all computed results: cross-validated predictive performance by information tier and model, calibration, subgroup fairness with bootstrap confidence intervals, capacity-limited targeting, and a robustness check. Four figures summarising the review funnel and benchmark results are included.

All bibliographic metadata derive from the open OpenAlex database (CC0). The benchmark dataset is redistributed under its original CC BY 4.0 licence with attribution. No personal or confidential data are included.