Detail Cantuman

Ketepatan waktu kelulusan mahasiswa merupakan indikator penting kualitas perguruan tinggi, namun beberapa mahasiswa di perguruan tinggi swasta masih menyelesaikan studinya melebihi waktu yang ditetapkan. Penelitian ini bertujuan untuk membandingkan performa beberapa algoritma klasifikasi machine learning Logistic Regression, K-Nearest Neighbors (KNN), Random Forest, dan XGBoost dalam memprediksi kelulusan tepat waktu mahasiswa, sekaligus menganalisis faktor-faktor yang paling berpengaruh terhadap prediksi tersebut. Data historis akademik alumni digunakan sebagai sumber informasi, dengan metodologi Cross Industry Standard for Data Mining (CRISP-DM) yang mencakup pengumpulan data, prapemrosesan, pemodelan, evaluasi, dan analisis kontribusi fitur. Penanganan ketidakseimbangan data dilakukan menggunakan SMOTE, dan tuning parameter diterapkan melalui GridSearch untuk meningkatkan performa model. Hasil penelitian menunjukkan bahwa Random Forest dengan GridSearch memiliki performa terbaik dengan akurasi 78,55% dan AUC 0,86, diikuti XGBoost yang juga kompetitif (AUC 0,87). Logistic Regression menunjukkan performa stabil, sementara KNN memiliki waktu eksekusi tercepat namun akurasi lebih rendah. Analisis kontribusi fitur mengidentifikasi Indeks Prestasi Kumulatif (IPK) dan Program Studi sebagai faktor dominan dalam memprediksi kelulusan tepat waktu, sedangkan fitur lain seperti SKS dan status SKS memberikan kontribusi tambahan. Penelitian ini memberikan dasar bagi pengambilan keputusan akademik berbasis data, membantu perguruan tinggi swasta merancang strategi pencegahan keterlambatan kelulusan, serta meningkatkan kualitas lulusan melalui pemilihan model prediksi yang paling optimal.

Timely graduation of students is a key indicator of the quality of higher education institutions. However, some students at private universities still complete their studies beyond the expected time frame. This study aims to compare the performance of several machine learning classification algorithms Logistic Regression, K-Nearest Neighbors (KNN), Random Forest, and XGBoost in predicting on-time graduation, while also analyzing the most influential factors affecting the prediction. Historical academic data from alumni were used as the main source of information, following the Cross Industry Standard Process for Data Mining (CRISP-DM) methodology, which includes data collection, preprocessing, modeling, evaluation, and feature contribution analysis. To address data imbalance, SMOTE was applied, and hyperparameter tuning was conducted using GridSearch to improve model performance. The results show that Random Forest with GridSearch achieved the best performance with an accuracy of 78.55% and an AUC of 0.86, followed by XGBoost, which also demonstrated competitive results (AUC 0.87). Logistic Regression showed stable performance, while KNN had the fastest execution time but lower accuracy. Feature contribution analysis identified Grade Point Average (GPA) and Study Program as dominant factors in predicting on-time graduation, with additional contributions from other features such as credit load (SKS) and SKS status. This study provides a data-driven foundation for academic decision-making, supporting private universities in designing strategies to prevent delayed graduation and improve graduate quality through optimal predictive model selection.