<![CDATA[Identifying which students may struggle in examinations early on is a critical challenge in vocational schools. This study aims to create and compare two machine learning models to predict the graduation status of Vocational High School (SMK) students majoring in Software and Game Development (PPLG). This prediction is based on their Competency Skills Test (UKK) scores. We used data from 310 students and tested two methods: Support Vector Machine (SVM) and Artificial Neural Network (ANN). The results are very clear: the SVM model performed exceptionally well, achieving an accuracy of 99%. SVM was able to recognize both 'Competent' and 'Not Yet Competent' students in a balanced manner. Conversely, the ANN model's performance was poor, with an accuracy of only 66%. This occurred because the ANN failed to learn and simply guessed that all students would pass. This research concludes that SVM is a highly effective method to be used as an early warning system. With this system, schools can more quickly assist students who are at risk of failing. SVM achieved 99% accuracy with perfect precision for the Competent class and full recall for the Not Yet Competent class. ROC-AUC and PR-AUC indicated excellent separability and strong minority-class detection. ANN achieved only 66% accuracy, predicting all samples as Competent. Learning curves revealed stagnation and failure to learn minority class patterns. Additional baseline models (Logistic Regression, Random Forest) were tested, with SVM outperforming all others consistently. Statistical significance testing using McNemar's test confirmed that SVM provides significantly better classification performance than ANN (p < 0.01).]]>
Copyrights © 2025
