Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2025
- Goals
- Record
- cc6ee8d5-6254-433c-bf60-48dd9eafa92a
- Published
- 2026-10-30
Universitas Dian Nuswantoro conducted research on early-stage diabetes detection using deep learning approaches for medical data. The activity involved UDINUS researchers Ramadhani, Talitha Olga; Shidik, Guruh Fajar; Novtagravina, Adelfia Milka; Nuraini, Aisyah; Widhiyanti, Erna Amalia; and Kusum, Edi Jaya. The research addressed diabetes as a significant global health risk and as one of the top causes of mortality worldwide. It focused on the importance of earlier identification of diabetes risk to avoid further complications.
The activity examined the effect of class imbalance in medical datasets, which can cause bias in prediction models. Within the ESDRP dataset, the research assessed five deep learning models: CNN, ANN, FFNN, MLP, and LSTM. Eight different resampling strategies were evaluated to mitigate class imbalance. The experimental results identified the ADASYN-ENN hybrid approach as providing the best performance. When combined with the CNN model, this approach achieved 98.70% accuracy, 100% precision, 96.15% recall, 98.04% F1-score, and a 99.85% AUC score.
The findings indicate that combining oversampling and undersampling was the optimal solution for diabetes detection with imbalanced data. Through this research activity, UDINUS addressed the stated goal of supporting health-related research by examining a method for improving diabetes detection under class imbalance conditions. The activity provides an institutional example of UDINUS research focused on medical data, deep learning models, and resampling strategies for diabetes detection.