Skip to content
SDGs Center UDINUS

Evidence record · UDINUS · SDGs

SMOTE-ENN Resampling to Optimize Diabetes Prediction in Imbalanced Data

Rektorat UDINUS 2025 Published 2026-10-30

Evidence record · UDINUS · SDGs

Unit
Rektorat UDINUS
Period
2025
Goals
Record
9bfa8d2a-665d-44fe-94d4-4a6626264ddd
Published
2026-10-30

Universitas Dian Nuswantoro conducted a study examining the role of resampling techniques in improving machine-learning model performance for detecting Diabetes Mellitus in imbalanced datasets. The activity involved Setiawan, Agustinus Eko; Rustad, Supriadi; Syukur, Abdul; Soeleman, Moch Arief; Akrom, Muhamad; and Setiawan, Andreas Wilson as participants from the institution.

The study used the BRFSS 2015 and PIMA Indian datasets. Five machine-learning algorithms were implemented on both datasets before and after resampling: KNN, RF, SVM, GB, and XGB. The resampling techniques examined were SMOTE, SMOTE-TOMEK, and SMOTE-ENN. The study found that all three resampling techniques improved the performance of all evaluated machine-learning models.

The strongest improvement for the BRFSS 2015 dataset was obtained from the combination of KNN and SMOTE-ENN, which achieved an accuracy of 0.97 and other evaluation metrics ranging from 0.96 to 0.99. For the PIMA Indian dataset, the same combination achieved a value of 1.0 for all evaluation metrics. The study also observed that resampling improved the correlation between each feature and the target, making it easier for the models to recognize data patterns.

For the goal of advancing health research, the study reported that, in imbalanced datasets, the role of resampling was more worthy of attention than the choice of algorithm. With SMOTE-ENN, the difference in performance between machine-learning models was at most 2 to 4 percent. UDINUS therefore presents this activity as research on resampling and machine-learning methods for Diabetes Mellitus prediction, with reported relevance to flexible application in the health sector.

Reference https://www.scopus.com/pages/publications/105009932678?origin=resultslist