PREDIKSI KELULUSAN TEPAT WAKTU MAHASISWA BERDASARKAN REKAM AKADEMIK MENGGUNAKAN KOMPARASI ALGORITMA RANDOM FOREST DAN XGBOOST

Abstract

Timely student graduation rates constitute a primary metric in academic program accreditation and higher education quality governance. This research engineers an early predictive modeling framework benchmarking Random Forest and Extreme Gradient Boosting (XGBoost) algorithms leveraging historical transcript data from semester 1 through semester 4 (semester GPAs, earned credit hours, and enrollment status). The dataset comprises 800 student records evaluated via 10-fold cross-validation. Empirical results confirm XGBoost superior performance, achieving 93.75% accuracy, 94.10% precision, 93.20% recall, and an AUC-ROC of 0.968, outstripping Random Forest (89.50% accuracy). Feature importance analysis revealed semester 3 GPA and semester 4 cumulative credits as the most pivotal predictors, enabling targeted early academic advising interventions.