A Literature Review of Empirical Studies on Predictive Learning Analytics for Student Success in Higher Education

Abstract

Predictive Learning Analytics (PLA) has emerged as a practical approach for using data from digital learning environments to better understand student progress and to support timely academic assistance. This research reviews recent work on PLA in higher education with a focus on how prediction is commonly framed and what factors repeatedly limit dependable use. The synthesis shows that PLA studies typically build models from routine academic and engagement-related information, aiming to flag students who may require additional support. At the same time, the literature raises consistent concerns about whether commonly used indicators capture meaningful learning process, whether model remains dependable when moved to new courses or cohorts, and whether results are presented in a way that can be responsibly interpreted and acted upon. Evidence on practical value of early-warning system is uneven, particularly when predictors are not embedded in clear support procedures. Overall, it argues that PLA is most defensible when models are validated within the context, communicated transparently, and implemented as part of student-centered support rather than as a standalone technical solution.