Related Experiment Videos
Interpretable diabetes risk prediction: a comparative study of tree-based algorithms using SHAP and LIME
Ruth Reátegui1,2, Lourdes Ramírez-Cerna3, Estefanía Bautista-Valarezo4
1Universidad Técnica Particular de Loja, Loja, 11-01-608, Ecuador. rmreategui@utpl.edu.ec.
Scientific Reports
|July 17, 2026
Summary
Machine learning models can predict diabetes risk in Ecuador. Key factors include blood glucose, family history, and physical activity, enabling early detection in low-income regions.
Area of Science:
- Public Health
- Biomedical Informatics
- Machine Learning
Background:
- The global prevalence of diabetes mellitus is rising, especially in low- and middle-income countries.
- Existing research on machine learning for diabetes risk prediction has limited data from Latin American populations.
- This study addresses the need for localized diabetes risk prediction models in Ecuador.
Purpose of the Study:
- To predict the risk of diabetes mellitus in the Ecuadorian population using tree-based machine learning algorithms.
- To identify key contributing factors for diabetes risk prediction in this specific demographic.
- To evaluate the performance and interpretability of different machine learning models.
Main Methods:
- Utilized a comprehensive dataset including sociodemographic, anthropometric, dietary, physical activity, clinical, and family history variables.
- Applied and compared four tree-based algorithms: Gradient Boosting, Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LGBM), and CatBoost.
- Employed explainable AI techniques (SHAP and LIME) to interpret model predictions and identify significant risk factors.
Main Results:
- Extreme Gradient Boosting (XGBoost) achieved the highest performance, with Area Under the Curve (AUC) values of 0.99 for training and 0.96 for testing.
- Key predictors identified include elevated blood glucose levels, family history of diabetes, abdominal circumference, age, visceral fat, low physical activity, waist-to-hip ratio, and muscle mass index.
- Explainable AI methods confirmed the importance of these clinical and lifestyle factors in diabetes risk.
Conclusions:
- Interpretable machine learning models, particularly XGBoost, show significant potential for accurate diabetes risk prediction in Ecuador.
- The identified key risk factors provide valuable insights for targeted early detection and prevention strategies.
- This approach is promising for improving diabetes management in low- and middle-income countries with limited resources.
