This project implements a two-stage machine learning pipeline for early diabetes detection using patient demographic and physiological data. The goal is to improve early diagnosis while reducing reliance on costly and time-consuming lab tests.
Early detection of diabetes is critical for preventing long-term complications. However, laboratory testing is expensive and time-intensive. This project uses non-invasive medical data to build an efficient machine learning pipeline that performs risk screening before final diagnosis.
| Stage | Algorithm | Purpose | Performance |
|---|---|---|---|
| Stage 1 | Logistic Regression | Initial screening | F1-score = 0.7773 |
| Stage 2 | Random Forest | Final classification | F1-score = 0.8696 |
- ✅ 95% detection sensitivity maintained
- ✅ Reduced unnecessary lab tests by 30%
- ✅ Improves healthcare resource allocation
- ✅ Balanced precision and recall for practical use
- Python
- scikit-learn
- pandas, numpy
- matplotlib
Part of my portfolio: haiiibin.github.io