<![CDATA[The cost of higher education in Indonesia varies greatly and often becomes a financial burden for students. Socio-economic factors such as parental income, occupation, number of dependents, vehicle ownership, and place of residence influence the determination of single tuition as regulated by the Ministry of Education Regulation No. 55 of 2013. This study aims to classify freshmen eligibility for single tuition relief using five machine learning models: RF, LR, KNN, SVM, and NB. The dataset contains 2000 rows of data with six socio-economic attributes divided into two classes: eligible and ineligible. The data were split into 80% training and 20% testing, and model performance was evaluated using accuracy, precision, recall, F1-score, and ROC-AUC. Results show that without SMOTE, all models suffer from severe majority-class bias, yielding critically low recall for the minority class SVM = 0.014; NB = 0.004. SMOTE significantly improves minority-class detection, with RF and SVM achieving the highest performance F1-scores of 0.820 and 0.801, and ROC-AUC of 0.966 and 0.990, respectively. SHAP analysis identifies Number of Dependents of Parents as the most influential predictor across all models, highlighting its central role in financial need assessment. These findings demonstrate that combining SMOTE with ensemble or margin-based models enhances classifiying fairness and sensitivity in educational support systems. The future work recommend expanding features to include behavioral, academic, and regional indicators, using multi-institutional data, and exploring deep learning or advanced resampling methods to enhance generalizability and robustness]]>
