Permutation-Based RFECV on Imbalanced Students’ Grade and Students’ Dropout Datasets

Irfan Pratama (Department of Information System, Universitas Mercu Buana Yogyakarta, Indonesia)
Albert Yakobus Chandra (Department of Information System, Universitas Mercu Buana Yogyakarta, Indonesia)
Putry Wahyu Setyaningsih (Department of Information System, Universitas Mercu Buana Yogyakarta, Indonesia)
Husna Sarirah Husin (School of Computer Science, Taylor’s University, Malaysia)
Odi Nurdiawan (Department of Informatics Management, STMIK-IKMI, Indonesia)

Jurnal Teknik Informatika (JUTIF)
Vol. 7 No. 4 (2026): JUTIF Volume 7, Number 4, August 2026

Predicting student academic performance is a critical task in EDM. To build predictive models, it is essential to identify the most informative features from LMS data. However, existing feature selection techniques exhibit significant drawbacks. Filter Methods evaluate features independently from the learning model, ignoring interactions and context. As a result, they may select redundant features and lack adaptability to specific classifiers. To enhance classification accuracy, this research proposes a permutation-based Recursive Feature Elimination with cross-validation (RFECV) feature selection method to strengthen machine learning models' student dataset classification, thereby eliminating bias from model-based feature importance calculations using the default method. The proposed method excels on the available educational datasets (Math, Por, Student Success, and Student dataset) in terms of F1-Score. Specifically, we achieved 80.59% on math dataset and 71.81% on student success dataset, both of which are multiclass datasets.