Heart Disease AI Prediction System

A machine learning system that predicts heart disease risk using clinical patient data.
Built with Python, Scikit-learn, Streamlit, FastAPI, and SHAP.

⚠️ This project is for educational purposes only and is NOT a medical diagnostic tool.

---

Features

• Multiple ML models (Random Forest, XGBoost, LightGBM, etc.)
• Automatic model selection
• Interactive Streamlit web app
• REST API using FastAPI
• Model explainability using SHAP
• Data visualization dashboard
• Risk probability scoring system

---

Dataset

Uses the UCI Heart Disease Dataset (processed Cleveland dataset):

Features include:

• Age
• Sex
• Chest pain type
• Blood pressure
• Cholesterol
• ECG results
• Max heart rate
• Exercise angina
• Oldpeak
• Thalassemia

Target:

• 0 = Low risk
• 1 = High risk

The training script downloads and cleans the dataset from the UCI repository if data/heart.csv is not already present.

---

Machine Learning Pipeline

1. Data preprocessing
2. Missing value handling
3. Encoding categorical variables
4. Feature scaling
5. Model training (10+ models)
6. Best model selection
7. Evaluation (Accuracy, ROC-AUC)
8. Model saving

---

Project Structure

heart-disease-ai/
│
├── app.py              # Streamlit UI
├── api.py              # FastAPI backend
├── train.py            # Model training
├── predict.py          # CLI prediction
├── explain.py          # SHAP explainability
├── dashboard.py        # Analytics dashboard
├── utils.py
├── config.py
│
├── models/
│   └── model.pkl
│
├── data/
│   └── heart.csv
│
└── requirements.txt

---

Run Locally

1. Install dependencies

pip install -r requirements.txt

---

2. Train model

python train.py

---

3. Run Streamlit App

streamlit run app.py

---

4. Run Dashboard

streamlit run dashboard.py

---

5. Run API

uvicorn api:app --reload

---

6. CLI Prediction

python predict.py

---

API Usage

Endpoint

POST /predict

Example Request

{
  "age": 58,
  "sex": 1,
  "cp": 2,
  "trestbps": 140,
  "chol": 220,
  "fbs": 0,
  "restecg": 1,
  "thalach": 150,
  "exang": 0,
  "oldpeak": 1.0,
  "slope": 2,
  "ca": 0,
  "thal": 2
}

Response

{
  "prediction": "High Risk",
  "probability": 0.91,
  "confidence": 0.91,
  "recommendation": "Consult a healthcare professional promptly.",
  "disclaimer": "Not a medical diagnosis."
}

---

Model Performance

• Accuracy: ~85%–95%
• ROC-AUC: ~0.88–0.96

(Depends on dataset split and model choice)

---

Tech Stack

• Python
• Pandas / NumPy
• Scikit-learn
• XGBoost / LightGBM / CatBoost
• Streamlit
• FastAPI
• SHAP
• Matplotlib / Seaborn

---

Disclaimer

This system is NOT a medical device. It is intended only for educational and research purposes.

Always consult a certified medical professional.

---

Author

Built as a machine learning project for educational purposes.

---

Future Improvements

• Deep learning model (TensorFlow / PyTorch)
• Real hospital dataset integration
• Mobile app version
• Docker deployment
• Cloud hosting (AWS / GCP)

requirements.txt

pandas
numpy
scikit-learn
matplotlib
seaborn
streamlit
fastapi
uvicorn
joblib
plotly
shap
xgboost
lightgbm
catboost

-Heart-Disease-Prediction-