<![CDATA[This study proposes a two-phase hybrid Educational Data Mining (EDM) framework that integrates K-Means clustering with supervised classification to predict students' final grade categories in Informatics within the Kurikulum Merdeka competency-based assessment system. The dataset consists of 281 Grade X students, all of whom achieved scores above the Minimum Achievement Criterion (KKTP = 75), with prediction focused on three grade categories: A (≥86), B (80–85), and C (75–79). Four summative assessment scores (S1, S7, S8, and S9) were used as input features. In the first phase, K-Means generated three clusters (Silhouette Score = 0.3198), and the resulting cluster labels were added as an additional feature. In the second phase, Random Forest and Logistic Regression were optimized using Grid Search with 5-Fold Stratified Cross-Validation, while SMOTE was employed to address class imbalance. The results show that Logistic Regression outperformed Random Forest, achieving a test accuracy of 59.65% and a Macro F1-Score of 0.5899, whereas Random Forest achieved 49.12% accuracy and a Macro F1-Score of 0.4799 and exhibited signs of overfitting. Feature importance analysis identified S7, S8, and S9 as the most influential predictors, while the cluster-derived feature contributed more strongly to Random Forest than to Logistic Regression. These findings suggest that well-regularized linear models may generalize better than ensemble methods on small datasets with narrow score distributions. The proposed framework is best positioned as a screening-support tool for early formative intervention in competency-based educational settings.]]>
Copyrights © 2026
