Interpretable Artificial Intelligence in Pedagogical Diagnostics and Math Performance Prediction: A Systematic Review of Applications in International Standardized Assessments

Artificial intelligence is already delivering tangible results in predicting student success. However, a critical challengeremains: most models function as “black boxes,” leaving teachers without a clear understanding of how decisionsare actually made. This review focuses on explainable artificial intelligence (XAI) methods for forecasting performance onstandardized assessments such as the SAT, GRE, and PISA. Specific tools–including SHAP, LIME, and attention mechanisms–are examined. The evidence indicates that XAI significantly enhances both assessment accuracy and transparency.Nevertheless, substantial challenges persist: explanations are often overly technical, and effective classroomintegration requires considerable preparatory work. This review systematizes practical use cases, identifies the strengthsof key methods, and maps concrete implementation barriers, offering value for stakeholders seeking to deploy AI responsiblyin educational diagnostics.