<![CDATA[Tuition fee reduction (Uang Kuliah Tunggal, UKT) is a key affordability instrument in Indonesian higher education, yet its assessment still relies largely on manual verification, producing decisions that are slow, inconsistent between reviewers, and difficult to justify to applicants. This study develops and evaluates an explainable machine learning decision support system that classifies the eligibility of tuition fee reduction applications and accompanies every prediction with a human-readable justification. Four tree-based ensemble models (Random Forest, XGBoost, CatBoost, and LightGBM) were trained on 5,000 application records from Politeknik Negeri Cilacap (2022–2024) described by socio-economic, academic, and administrative attributes, and benchmarked against a Logistic Regression baseline using hold-out testing, 10-fold cross-validation, and the Friedman test. SHAP (SHapley Additive exPlanations) was then applied to the best-performing model to quantify global feature contributions and generate per-application explanations. XGBoost achieved the highest accuracy (0.822) and F1-score (0.850), while Random Forest obtained the highest ROC-AUC ([≈0.89]); however, all models remained within approximately 2.6 percentage points of accuracy, and the Logistic Regression baseline attained the highest precision (0.838). The SHAP analysis revealed that number of dependents and document completeness dominated the decision function, far exceeding parental income, indicating that the historical decision process encoded an administrative gate alongside genuine economic need. These findings support deploying the system as a human-in-the-loop screening aid rather than an autonomous decision maker, and demonstrate that explainability is essential for accountable financial aid decisions.]]>
