Prediction of thirty-day hospital readmission in type 2 diabetes mellitus patients Using hybrid recursive feature elimination and random forest
Computational Design: Journal of Modern Informatics · 30.09.2026 · 1 (2) · 158-167
DOI: 10.35730/705fpa09 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computational Design: Journal of Modern Informatics, online erschienen am 30.09.2026. Verfasst von Rosi Diana und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-SA 4.0).
Themen: Diabetes mellitus
Abstract (Originalsprache)
Hospital readmission within thirty days among type 2 diabetes mellitus patients represents a critical challenge in clinical management, leading to elevated healthcare expenditures and increased morbidity rates. Electronic health records provide abundant clinical variables to construct predictive models; however, these high-dimensional datasets frequently contain redundant and noisy features that degrade classification performance and increase computational complexity. This study proposes a hybrid machine learning framework that integrates Recursive Feature Elimination (RFE) and Random Forest (RF) to identify optimal clinical determinants and accurately predict thirty-day hospital readmission. The experimental evaluation was conducted using the Diabetes 130-US Hospitals dataset, comprising 101,766 patient encounters and 47 initial attributes. Following data preprocessing, missing value imputation, and class balancing using the Synthetic Minority Over-sampling Technique (SMOTE), the RFE algorithm iteratively eliminated weak attributes, successfully reducing the feature space from 47 variables to 8 core clinical indicators. The experimental results demonstrate that the hybrid RFE-RF model outperforms baseline classifiers, achieving an accuracy of 0.864, a recall (sensitivity) of 0.882, a precision of 0.849, an F1-score of 0.865, and an Area Under the Receiver Operating Characteristic Curve (AUC-ROC) of 0.912. Clinical analysis reveals that the number of prior inpatient visits, discharge disposition, glycosylated hemoglobin (HbA1c) test results, insulin therapy adjustments, and time in hospital are the most influential predictors of early readmission. These findings confirm that combining wrapper-based feature selection with ensemble tree modeling significantly improves predictive sensitivity and offers a reliable clinical decision support tool for proactive post-discharge intervention
Zitieren
APA
Diana, R., .B, E. H., & Ediana, I. R. (2026). Prediction of thirty-day hospital readmission in type 2 diabetes mellitus patients Using hybrid recursive feature elimination and random forest. Computational Design: Journal of Modern Informatics, 1(2), 158-167. https://doi.org/10.35730/705fpa09
Vancouver
Diana R, .B EH, Ediana IR. Prediction of thirty-day hospital readmission in type 2 diabetes mellitus patients Using hybrid recursive feature elimination and random forest. Computational Design: Journal of Modern Informatics. 2026;1(2):158-167. doi:10.35730/705fpa09
BibTeX
@article{Diana2026,
author = {Diana, Rosi and .B, Erwin Herian and Ediana, Irwan Rahmadi},
title = {Prediction of thirty-day hospital readmission in type 2 diabetes mellitus patients Using hybrid recursive feature elimination and random forest},
journal = {Computational Design: Journal of Modern Informatics},
year = {2026},
volume = {1},
number = {2},
pages = {158--167},
doi = {10.35730/705fpa09}
}
EUVIMED