manpreet-sangha/predict_coronary_heart_disease

Predict the coronary heart disease (chd: 1/0) for males in a heart-disease high-risk region of the Western Cape, South Africa # Predicting Coronary Heart Disease A machine learning pipeline to predict coronary heart disease (CHD) in males from a high-risk region of the Western Cape, South Africa. **Module:** SMM748 Machine Learning For Quantitative Professionals — Individual Coursework **Live App:** predict-coronary-heart-dise… ## Dataset **Source:** `input_data/heart-disease.csv` — 462 patients, 9 clinical features, binary target (`chd`: 1 = disease, 0 = no disease). | Feature | Description | |---|---| | `sbp` | Systolic blood pressure | | `tobacco` | Cumulative tobacco consumption (kg) | | `ldl` | Low density lipoprotein cholesterol | | `adiposity` | Adiposity index | | `famhist` | Family history of heart disease (Present/Absent) | | `typea` | Type-A behaviour score | | `obesity` | Obesity index | | `alcohol` | Current alcohol consumption | | `age` | Age in years | --- ## Key Results - **Ridge Logistic Regression** achieves the highest test accuracy (**0.753**) across all 12 classifiers. - AdaBoost, QDA, and SVM (RBF) tie at 0.742. - Log1p preprocessing benefits generative classifiers (QDA +3.2%) but does not affect Ridge LR's result. - `age`, `age_famhist`, and `ldl` are consistently the strongest CHD predictors (confirmed by ridge coefficients, feature importance, and SHAP analysis). --- ## Project Structure ``` predict_coronary_heart_disease/ ├── chd_main.py # Main entry point — runs full pipeline ├── config.py # Project-wide constants (features, paths, seeds) ├── requirements.txt # Python dependencies ├── input_data/ │ └── heart-disease.csv ├── feature_engineering/ # Derived interaction features │ ├── fe.py # age_tobacco, age_famhist │ ├── fe_age_tobacco.py │ └── fe_age_famhist.py ├── exploratory_data_analysis/ # Section 1: EDA │ ├── eda.py # Orchestrator │ ├─ …