Abstract
<title>Abstract</title> <p>Mathematics underachievement remains one of the most persistent challenges facing basic and higher education systems worldwide, and it is increasingly addressed through learning analytics (LA) and early warning systems (EWS) that convert routinely collected academic data into actionable risk indicators. This review synthesises evidence from thirty peer-reviewed and institutional studies published mainly between 2019 and 2026 to evaluate how LA-EWS programmes have been designed, implemented and assessed for mathematics achievement across diverse national contexts, including Chile, Germany, the United Kingdom, Estonia, Mexico, Türkiye and several cross-national assessment regimes. The review adopts a structured narrative-synthesis methodology informed by systematic review conventions, comparing predictor variables, modelling approaches, reported discriminative performance and equity outcomes. Findings indicate that discrimination, expressed as the area under the receiver operating characteristic curve, typically ranges between 0.70 and 0.93, yet operational usefulness depends heavily on prevalence stability, threshold governance and the extent to which indicators capture domain-specific mathematical misconceptions rather than generic engagement signals. Persistent gaps are identified in algorithmic fairness, cross-country comparability, and the integration of conceptual diagnostic evidence, particularly around student difficulties with functions and graphical representation. The review further considers how emerging acceptance of generative artificial intelligence tools among tertiary populations is likely to reshape future analytics adoption. The paper concludes with a comparative framework and policy-relevant recommendations for designing mathematics-specific, equity-aware early warning programmes suitable for cross-country evaluation.</p>