<![CDATA[This study develops a classification pipeline for assessing student aid eligibility using administrative school data with feature interpretation to support accountable preliminary review. The dataset consists of 1,242 anonymized student records from one senior high school in Ambarawa, Central Java. Prior to modeling, direct identity attributes and label-leakage-prone features, such as eligibility reasons and aid card numbers, were removed. Five classification algorithms were evaluated using an 80:20 stratified train-test split and five-fold cross-validation. Random Forest was selected as the main model because it provided a balance between predictive performance and feature interpretability, achieving an accuracy of 0.6988, an F1-score of 0.6939, and a ROC-AUC of 0.7736. Gradient Boosting achieved a higher ROC-AUC of 0.7867; however, its performance gain was relatively small, while the feature importance results from Random Forest were easier to communicate to non-technical school staff in an implementation context. Feature interpretation indicates that father’s income was the most influential predictor, with an importance score of 0.1894, whereas distance and transportation mode provided secondary accessibility signals. The proposed pipeline is not intended as an automated decision-making mechanism, but as a tool for generating a priority list of candidates requiring human verification and local retraining before being adapted by other schools.]]>
Copyrights © 2026
