← Back to work
Machine Learning & Clinical Research

Type 2 Diabetes Prediction Using Vascular Stiffness Parameters

📄 Presented at ISPACS 2025 (International Symposium on Intelligent Signal Processing and Communication Systems)
Random ForestXGBoostSVMScikit-LearnLOOCV

Problem statement

Type 2 Diabetes Mellitus (T2DM) is a major global health concern associated with severe cardiovascular complications, including increased arterial stiffness. An effective, non-invasive screening method is critical for early risk assessment before complications develop.

My contribution

Comparative Study

Trained and validated four models — Logistic Regression, SVM, Random Forest, and XGBoost — on a dataset of 69 Indonesian subjects (21 diagnosed with diabetes).

Robust Validation

Applied a Leave-One-Out Cross-Validation (LOOCV) strategy to get reliable performance estimates on a small clinical dataset.

Feature Engineering

Used 13 non-invasive predictors spanning demographic/anthropometric (Age, BMI, Vascular Age) and hemodynamic/vascular parameters (ABI, baPWV, Aortic & Peripheral Compliance).

Technical stack

Random ForestXGBoostSVMLogistic RegressionScikit-LearnLOOCV

Results & impact

  • Random Forest achieved the best performance: 85.5% overall accuracy, 0.921 AUC, and a 0.74 F1-score for the diabetic class.
  • Demonstrated that non-invasive vascular parameters are a viable basis for early diabetes screening.
  • Positioned the approach as a potential clinical decision support tool.

Have a similar challenge you want to solve?

Start a project discussion