<![CDATA[Explainable artificial intelligence (XAI) is increasingly used to support prediction, assessment, feedback, and instructional decisions in mathematics education, yet evidence about how explanation affects transparency, trust, and learning remains fragmented. This systematic literature review synthesized evidence on explainability in mathematics and closely related STEM education, with particular attention to the under-studied middle-school learner. The review followed PRISMA 2020 and used a structured TITLE-ABS-KEY search of Scopus. The search returned 646 records; 550 were excluded during title/abstract screening, 96 reports were assessed at full text, and 10 studies were retained for qualitative thematic synthesis. Three themes emerged: interpretable prediction of performance and at-risk learners; transparency, trust, and adoption; and explainable scaffolding and feedback. The evidence base is dominated by 2025 studies and by educator- or institution-facing prediction, while direct learner-facing explanation and experimentally tested effects on achievement remain scarce. SHAP, feature-importance methods, neuro-fuzzy models, and explanatory feedback systems improve access to model reasoning, but explanation quality alone has not been shown to cause higher achievement. The review therefore distinguishes technical interpretability from pedagogical usefulness and identifies validated measurement needs for teacher decision quality, student self-regulation, calibrated AI trust, and equitable treatment. Because the archived screening and appraisal logs were not preserved, retrospective inter-rater coefficients and aggregate quality scores could not be reconstructed without inventing data; this limitation is reported explicitly. Future studies should use dual-reviewer screening, design-appropriate critical appraisal, classroom-embedded experiments, and age-appropriate explanations.]]>
Copyrights © 2025
