EXPLAINABLE ARTIFICIAL INTELLIGENCE (XAI) FOR INTERPRETING STUDENT PERFORMANCE PREDICTION IN STEM AND COMPUTER SCIENCE: A SYSTEMATIC LITERATURE REVIEW

Abstract

Educational Data Mining (EDM) enables the early identification of learning difficulties through the prediction of students' academic performance. However, machine learning models often operate as "black boxes," making their decision-making processes difficult to interpret. Explainable Artificial Intelligence (XAI) addresses this challenge by providing insights into why these models generate specific predictions. This article presents a systematic literature review of research published over the past five years concerning the application of Explainable Artificial Intelligence (XAI) to predict student performance in computer science and STEM education. The findings indicate that the most frequently used features were students' behavioral data and academic records, while the primary prediction tasks focused on grade prediction and identifying students at risk of academic failure. The study also analyzed the application domains of XAI, identifying global feature importance analysis, the explanation of individual predictions, and support for pedagogical interventions as the most prevalent areas. Furthermore, SHapley Additive exPlanations (SHAP) emerged as the most widely used XAI method, applied predominantly for global model interpretation, with limited use for explaining individual predictions. Finally, a significant research gap was identified regarding the use of Explainable Artificial Intelligence (XAI) for course optimization, customized visualizations, and the generation of personalized recommendations. Addressing these gaps could enable educators to provide data-driven, personalized support to improve individual student outcomes.