The Application of XGBoost to Classify Students’ Interests in Computer Science at SMP Negeri 30 in South Bengkulu

Abstract

Student interest in Informatics is still commonly identified through teacher observation, creating a need for a more objective data-driven approach. This study develops an XGBoost classification model and explains the factors underlying its decisions using SHAP. A quantitative data-analytics design followed the CRISP-DM stages using 27 student records and seven features: assignment scores, daily tests, practical scores, midterm scores, final scores, attendance, and attitude. The dataset was split into 80% training and 20% testing data. On six held-out cases, the model correctly classified two students as Interested and four as Not Interested, producing 100% Accuracy, Precision, Recall, and F1-Score, with an AUC of 1.00. SHAP analysis identified daily-test scores as the strongest contributor (1.1179), followed by assignment, final-exam, practical, and midterm scores. These findings indicate that XGBoost combined with SHAP can support measurable and transparent identification of student interest, although the limited dataset requires caution when generalizing the performance.