<![CDATA[Mathematics achievement during the K-12 years strongly shapes later educational and economic opportunity, making early identification of struggling learners a priority. Although numerous predictive models have been developed, previous findings remain fragmented regarding predictive algorithms, influential predictors, and educational implications. Therefore, this review systematically synthesizes methodological trends, predictive performance, and research gaps across K–12 mathematics education. Only studies meeting predefined methodological quality criteria were included. Therefore, this study conducts a PRISMA 2020-based systematic literature review to synthesize predictive methodologies, influential predictors, methodological quality, and educational implications across K–12 mathematics education. Although digital educational data have become increasingly available, previous predictive studies have produced inconsistent findings due to differences in datasets, machine learning algorithms, validation strategies, and educational contexts. Consequently, researchers still lack a comprehensive understanding of which predictive approaches consistently perform well across K–12 mathematics education. From 583 Scopus records, ten studies (2019, 2025) qualified based on eligibility and quality criteria and were synthesised thematically. Tree-based ensemble and boosting methods, which have been supported by administrative, large-scale-assessment, and digital-process data, are the primary means of prediction; prior attainment, socio-economic context, and self-regulated learning are the most significant predictors. Instead of accuracy-centred prediction, the field is moving toward explainable, fairness-aware, and pedagogically embedded analytics. The review charts a scattered evidence base and urges longitudinal, primary-grade, externally validated, and equity-focused research so that prediction can responsibly enhance mathematics learning.]]>
