Predicting Educational Outcomes Using Artificial Intelligence

Abstract

In the context of the digital transformation of education and the rapid expansion of online learning systems, the problem of early prediction of educational outcomes has become strategically important. Modern approaches to educational data analysis, such as Educational Data Mining (EDM) and Learning Analytics, enable the development of highly accurate predictive models. However, many of these models operate as “black boxes,” limiting the transparency of decision-making processes and reducing trust among educators and educational administrators. In this regard, Explainable Artificial Intelligence (XAI) is increasingly recognized as a key instrument for improving the interpretability, ethical compliance, and practical applicability of machine learning models in education. The objective of this study is to develop and interpret a predictive model for student academic performance using XAI methodologies. The research applies a comparative analysis of Random Forest, XGBoost, and neural network algorithms, combined with interpretability techniques including SHAP, LIME, and Partial Dependence Plots (PDP). The study is based on a comprehensive educational dataset that includes academic performance indicators, demographic characteristics, and digital behavioral metrics extracted from Learning Management System (LMS) platforms. The findings indicate that gradient boosting models provide the highest predictive accuracy (ROC-AUC > 0.89). However, the integration of SHAP analysis enables the identification of key predictors of academic performance, including LMS interaction intensity, prior academic achievements, and behavioral engagement patterns. The interpretability analysis also reveals nonlinear relationships between study time investment and final academic performance. The practical significance of the study lies in establishing a methodological foundation for the development of transparent early-warning systems for academic risk detection, which can support evidence-based decision-making in educational institutions. The results further highlight the importance of integrating algorithmic fairness and AI ethics principles into educational analytics systems.