<![CDATA[Al-Manshurin Ar-Rikzan Jepara Islamic Boarding School faces challenges in the early detection of potential declines in student achievement because the evaluation process still relies on manual recording. This study aims to design an interactive website-based Early Warning System to predict the final performance predicate of students by integrating academic data and non-academic behavior (mutabaah yaumiyah). The methodology used refers to the Cross-Industry Standard Process for Data Mining (CRISP-DM) framework. This study compares three classification algorithms: Decision Tree (C4.5), Naïve Bayes, and K-Nearest Neighbor (K-NN). To address class imbalance in the 210 original data records, the Synthetic Minority Over-sampling Technique (SMOTE) was applied to generate 450 balanced records, which were then evaluated using 10-Fold Cross Validation. The test results show that the Decision Tree (C4.5) algorithm produces the most superior performance with an Accuracy of 97.62%, Precision of 97.96%, and Recall of 97.62%. This superiority is driven by the decision tree structure's ability to map non-linear conditional rules relevant to the pesantren's absolute rules. As an applicative output, the best model is extracted into a Streamlit-based dashboard equipped with expert recommendations (feature importances). This system automatically highlights the variables contributing the highest risk, enabling administrators to formulate accurate intervention and mentoring steps before the semester evaluation ends.]]>
Copyrights © 2026
