The integration of artificial intelligence (AI) in the field of education offers promising opportunities for personalized learning and the prediction of student performance. However, the adoption of these technologies is often hindered by the lack of transparency and explainability of AI models, particularly those considered as “black boxes.” This article explores the use of explainable AI (XAI) techniques, specifically the SHAP (SHapley Additive exPlanations) method, to improve the interpretability of machine learning models in an educational context. We apply the XGBoost (eXtreme Gradient Boosting) algorithm to predict student performance in mathematics using the Student Performance dataset from the UCI Machine Learning Repository, and then use SHAP to explain the model’s predictions. The results show that previous grades (G1 and G2, i.e., the first and second evaluations) are the most influential factors, while absences have a significant negative impact. Additionally, features such as study time and family relationships contribute positively, though to a modest extent. SHAP visualizations, such as Summary Plots, Waterfall Plots, and Dependence Plots, reveal complex and non-linear relationships between features, highlighting the importance of a multifactorial approach to improving student performance. This study demonstrates that explainable AI techniques, such as SHAP, can not only improve the transparency of predictive models but also provide valuable insights to guide targeted and equitable educational interventions.