<![CDATA[This study develops and evaluates machine learning models to predict high school graduation outcomes and identify at-risk students for early intervention. Using a quantitative approach, data from 1,017 students across three public high schools were analyzed, encompassing academic performance (average yearly scores), behavioral factors (attendance rates and extracurricular participation), and socio-economic background (proxied by parental occupation). A comparative modeling strategy was applied, beginning with a Decision Tree baseline and advancing to a Stacking Ensemble model that integrated three heterogeneous base learners—Support Vector Machine (SVM), k-Nearest Neighbors (k-NN), and Decision Tree—combined through a Logistic Regression meta-model. Both models were optimized using GridSearchCV and adjusted for class imbalance between graduates (93.4%) and at-risk students (6.6%). The results showed that academic variables, particularly third-year average scores (mean = 82.6, SD = 6.4) and attendance rate (mean = 94.3%), were the strongest predictors of graduation, while socio-economic indicators had minimal impact. The Stacking Ensemble achieved a notable improvement over the Decision Tree, reaching an accuracy of 99.6%, precision of 0.909, recall of 1.000, F1-score of 0.952, and AUC of 1.000, compared to the baseline accuracy of 94.9% (F1-score = 0.519, AUC = 0.83). These findings indicate the superior predictive capability of the ensemble model in identifying students at risk of non-graduation. The study’s novelty lies in combining interpretable and high-performance models to construct a practical early-warning framework that can guide educators and policymakers in targeted academic interventions. However, the near-perfect metrics also suggest potential overfitting, emphasizing the need for validation using external datasets before broader application. Overall, this research contributes a robust, data-driven methodology for improving student retention through predictive analytics in educational settings.]]>
Copyrights © 2026
