Explainable Artificial Intelligence for Accurate and Interpretable Student Selection Prediction Without Aggregate Scores Using SKD Examination Data

Student selection processes are often based on aggregate scores, which can obscure the contribution of individual assessment components and limit interpretability. This study proposes an explainable machine learning approach to predict student selection outcomes using SKD examination data. An XGBoost model was developed and evaluated under two configurations, including a baseline model with aggregate score and a proposed model without it. The results show that the proposed model achieves high predictive performance, with an accuracy of 95.84% and ROC-AUC of 0.9935, demonstrating that individual test components provide sufficient information for reliable prediction. To enhance transparency, SHAP was applied to provide both global and local explanations of model decisions. The analysis reveals that cognitive test components, particularly TIU and TWK, are the most influential factors, while non-academic variables such as formation and location also contribute to the prediction. These findings highlight the importance of interpretability in high-stakes decision-making and suggest that accurate and transparent student selection systems can be developed without relying on aggregate scores. The proposed approach provides a robust framework for supporting fair, interpretable, and data-driven selection policies.