Machine Learning for Early Intervention: A Quantitative Systematic Review of Predictive Models for Undergraduate Mathematics Performance

Abstract This systematic review synthesizes current literature on the use of machine learning (ML) models to predict undergraduate mathematics performance and support early intervention in resource-constrained higher education environments (RCCEs). A systematic search of five academic databases was conducted in accordance with PRISMA 2020 guidelines, resulting in 19 empirical studies being included. The findings reveal that ensemble techniques such as Random Forest and XGBoost demonstrate strong predictive performance (\(\:78\--94\%\) accuracy) and AUC values ranging from \(\:0.84\) to \(\:0.96\) in high-resource contexts, using predictors such as prior academic achievement and Learning Management System (LMS) usage data. However, the straightforward application of these models to RCCEs, such as Ethiopian universities, is challenged by infrastructural limitations and data scarcity. Effective implementation requires a context-sensitive strategy emphasizing (1) interpretable and transparent models based on readily available data, (2) substantial preliminary investment in foundational academic support systems prior to predictive analytics deployment, and (3) the adoption of a rigorous ethical framework to mitigate algorithmic bias. Overall, this review highlights the need to shift research and practice from a narrow focus on technical model performance toward contextually relevant, ethically responsible, and intervention-driven applications.