PREDICTING STUDENTS AT RISK OF DROPPING OUT USING XGBOOST BASED ON ACADEMIC DATA

Abstract

Abstract: Student dropout has become a significant challenge for higher education institutions because it negatively affects academic performance, institutional reputation, and resource allocation. Early identification of students at risk of dropping out enables universities to implement timely interventions and improve student retention. This study proposes a predictive model for identifying students at risk of dropping out using the Extreme Gradient Boosting (XGBoost) algorithm based on academic data. The dataset consists of student academic records, including grade point average (GPA), course completion rate, attendance, accumulated credits, failed courses, and semester performance. The data were preprocessed through data cleaning, feature selection, and normalization to improve model performance. The dataset was then divided into training and testing sets using an 80:20 ratio. The XGBoost model was trained and evaluated using accuracy, precision, recall, F1-score, and Area Under the Receiver Operating Characteristic Curve (AUC-ROC). The experimental results demonstrate that XGBoost effectively captures complex relationships among academic variables and provides high predictive performance in identifying students with dropout risk. Feature importance analysis further reveals that GPA, accumulated credits, attendance rate, and the number of failed courses are the most influential factors affecting student dropout. The proposed approach offers a practical decision-support tool for higher education institutions by enabling early detection of at-risk students and facilitating targeted academic support programs. The findings contribute to the application of machine learning in educational data mining and learning analytics, providing valuable insights for improving student retention strategies and academic success. Future research may integrate non-academic factors such as socioeconomic background, psychological characteristics, and student engagement to further enhance prediction accuracy and model generalizability across different educational institutions.

Keywords: Student Dropout Prediction; XGBoost; Machine Learning; Educational Data Mining; Learning Analytics.

Abstrak: Putus kuliah merupakan tantangan signifikan bagi institusi pendidikan tinggi karena berdampak negatif terhadap kinerja akademik, reputasi institusi, dan alokasi sumber daya. Identifikasi dini terhadap mahasiswa yang berisiko putus kuliah memungkinkan universitas untuk menerapkan intervensi yang tepat waktu dan meningkatkan retensi mahasiswa. Penelitian ini mengusulkan model prediktif untuk mengidentifikasi mahasiswa yang berisiko putus kuliah menggunakan algoritma Extreme Gradient Boosting (XGBoost) berdasarkan data akademik. Kumpulan data mencakup rekam jejak akademik mahasiswa, termasuk Indeks Prestasi Kumulatif (IPK), tingkat penyelesaian mata kuliah, kehadiran, akumulasi kredit, mata kuliah yang gagal, dan kinerja semester. Data diproses melalui tahapan pembersihan data, seleksi fitur, dan normalisasi untuk meningkatkan kinerja model. Selanjutnya, kumpulan data dibagi menjadi set pelatihan dan pengujian dengan rasio 80:20. Model XGBoost dilatih dan dievaluasi menggunakan metrik akurasi, presisi, recall, skor-F1, dan Area Under the Receiver Operating Characteristic Curve (AUC-ROC). Hasil eksperimen menunjukkan bahwa XGBoost secara efektif menangkap hubungan kompleks antarvariabel akademik dan memberikan kinerja prediksi yang tinggi dalam mengidentifikasi mahasiswa dengan risiko putus kuliah. Analisis kepentingan fitur mengungkapkan bahwa IPK, akumulasi kredit, tingkat kehadiran, dan jumlah mata kuliah yang gagal merupakan faktor paling berpengaruh terhadap risiko putus kuliah mahasiswa. Pendekatan yang diusulkan ini menawarkan alat pendukung keputusan yang praktis bagi institusi pendidikan tinggi dengan memungkinkan deteksi dini mahasiswa berisiko serta memfasilitasi program dukungan akademik yang terarah. Temuan ini berkontribusi pada penerapan machine learning dalam penambangan data pendidikan dan analitik pembelajaran, serta memberikan wawasan berharga untuk meningkatkan strategi retensi mahasiswa dan keberhasilan akademik. Penelitian di masa mendatang dapat mengintegrasikan faktor non-akademik—seperti latar belakang sosial-ekonomi, karakteristik psikologis, dan keterlibatan mahasiswa—untuk lebih meningkatkan akurasi prediksi dan kemampuan generalisasi model di berbagai institusi pendidikan.

Kata kunci: Prediksi Putus Kuliah Mahasiswa; XGBoost; Machine Learning; Penambangan Data Pendidikan; Analitik Pembelajaran.