Target variable distribution highlighting significant class imbalance in the CDC Diabetes Health Indicators dataset
Zoom In
Pairwise correlations revealing relationships between health indicators, demographics, and diabetes prevalence
Zoom In
Feature Importance
Zoom In
Model comparison with confusion matrices across multiple classifiers
Zoom In

Project information

  • Category: Machine Learning / Healthcare Analytics
  • Project duration: Sep '24 - Dec '24
  • Team size: 3
  • Dataset: CDC Diabetes Health Indicators (253,680 samples)
  • Report URL: Detailed Report

Diabetes Risk Prediction

Technology(s) Used: Python, Scikit-learn, Pandas, GridSearchCV, Random Forest, SVM

  • Data Engineering: Cleaned and processed the massive CDC Diabetes Health Indicators dataset. Addressed significant class imbalance using oversampling techniques to ensure the minority (diabetes/prediabetes) class was properly represented.
  • Feature Selection: Conducted multi-collinearity analysis and used Random Forest Gini importance to distill 21 variables down to 7 key health indicators, optimizing model efficiency.
  • Model Development: Implemented and compared Logistic Regression, SVM, Random Forest, and KNN. Utilized GridSearchCV for hyperparameter tuning and cross-validation to prevent overfitting.
  • Performance: Achieved an 88.79% accuracy and 89.24% precision. The model provides a reliable framework for early screening based on behavioral and health metrics.