<![CDATA[Student graduation is an important indicator in accreditation and serves as part of quality management strategies in higher education. Therefore, early prediction of student graduation is necessary to improve the effectiveness of data-driven admission decision-making. Differences in student graduation rates are influenced by a combination of academic, demographic, economic, and family factors. This study applies the Stacking Ensemble Learning method by combining Random Forest, K-Nearest Neighbors, and Support Vector Machine, with XGBoost serving as the meta-learner. The dataset used integrates student admission records and graduation status reports from the NeoFeeder PDDikti system, covering 16 academic and non-academic feature variables. The model was evaluated using accuracy, precision, recall, F1-score, and Area Under the Curve (AUC). The results show that the stacking ensemble model outperformed single models, achieving 82% accuracy, a weighted F1-score of 80%, and an AUC of 87.15% on the test data. These findings contribute both the selected feature set and the implementation of an ensemble model for building a machine learning–based prediction system, particularly in addressing data imbalance and improving classification accuracy.]]>
Copyrights © 2025
