Predicting Diabetic Retinopathy and Nephropathy Complications Using Machine Learning Techniques
Diabetes and its complications, especially Diabetic Retinopathy (DR) and Diabetic Nephropathy (DN) is a big challenge to the global healthcare system and needs accurate predictive models to help in early diagnosis and intervention. In this study we used a dataset from a reputed medical center in Ind...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10969779/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Diabetes and its complications, especially Diabetic Retinopathy (DR) and Diabetic Nephropathy (DN) is a big challenge to the global healthcare system and needs accurate predictive models to help in early diagnosis and intervention. In this study we used a dataset from a reputed medical center in India with 767 patient records and 22 attributes including demographic details, clinical markers and treatment plans. We used a suite of advanced machine learning algorithms—Random Forest, XGBoost, LightGBM, CatBoost, Neural Networks and ensemble approaches like Voting and Stacking Classifiers to see their performance on original, oversampled and undersampled datasets. Through feature engineering, sampling strategies and hyperparameter tuning the models performed well on all the datasets. Surprisingly the models performed well even on the original imbalanced dataset which can be attributed to the power of the models and hyperparameter tuning. Ensemble methods like Voting and Stacking Classifiers performed better and achieved near perfect metrics (AUC = 1.0) in oversampled datasets. Hyperparameter tuning further improved the performance, reduced RMSE and log loss and increased accuracy and recall in all the configurations. This shows the importance of model optimization in real world clinical datasets which are imbalanced and noisy. This paper shows the possibility of machine learning based frameworks in diabetic complication management by predicting accurately and in time. These models can be integrated into clinical decision support systems (CDSS) to give insights to clinicians, improve patient outcomes through personalized interventions and optimize resource allocation. Future work will be to validate this on different populations, include longitudinal patient data and integrate real time electronic health records (EHR) for deployment in hospitals. |
|---|---|
| ISSN: | 2169-3536 |