Heart Disease Risk Prediction System

This project is an end-to-end machine learning system that predicts the risk of heart disease using clinical features. It is built using the UCI Cleveland Heart Disease dataset and deployed as an interactive Streamlit web application.

⚠️ This project is a research prototype and not a medical diagnostic tool.

---

Problem Statement

Early detection of heart disease is critical for preventive healthcare. This project aims to estimate the probability of heart disease based on patient clinical attributes using machine learning.

---

Dataset

• Source: UCI Cleveland Heart Disease Dataset
• Samples: 304 patients
• Target:
  • 0 → No heart disease
  • 1 → Heart disease present

Only the Cleveland dataset was used to avoid data leakage and corrupted labels present in other variants.

---

Machine Learning Pipeline

1. Data filtering (Cleveland-only)
2. Target binarization
3. Feature selection & cleanup
4. Encoding:
   • Binary: sex, fbs, exang
   • One-hot: chest pain (cp), restecg
5. Train/test split (stratified)
6. Models:
   • Logistic Regression (baseline)
   • XGBoost (final model)
7. Probability calibration (Isotonic Regression)
8. Model explainability using SHAP
9. Deployment using Streamlit

---

Model Performance (Test Set)

Model	Accuracy	F1 Score	ROC-AUC
Logistic Regression	0.87	0.85	0.92
XGBoost (Calibrated)	0.89	0.88	0.92

Calibration improved probability reliability (Brier score: 0.094).

---

Key Insights (SHAP)

Top predictive features:

• Sex
• Oldpeak (ST depression)
• Maximum heart rate (thalach)
• Age
• Chest pain type

---

Tech Stack

• Python
• scikit-learn
• XGBoost
• SHAP
• Streamlit
• pandas, numpy
• joblib

---

How to Run Locally

pip install -r requirements.txt
streamlit run app/app.py