Heart Disease ML Pipeline & Streamlit App

A complete, end-to-end machine learning workflow on the UCI Heart Disease dataset, including preprocessing, feature engineering, model training, evaluation, and an interactive Streamlit app for single and batch inference.

Project Structure

Heart_Disease_Project/
├─ data/
│  ├─ heart_disease.csv         # Raw dataset
│  ├─ cleaned_X.csv             # Cleaned, model-ready features (defines input schema)
│  └─ clean_y.csv               # Target labels used during training
├─ models/
│  └─ final_model.pkl           # Trained scikit-learn model artifact
├─ notebooks/                   # EDA, preprocessing, modeling, tuning
│  ├─ 01_data_preprocessing.ipynb
│  ├─ 02_pca_analysis.ipynb
│  ├─ 03_feature_selection.ipynb
│  ├─ 04_supervised_learning.ipynb
│  ├─ 05_unsupervised_learning.ipynb
│  └─ 06_hyperparameter_tuning.ipynb
├─ results/
│  └─ evaluation_metrics.txt    # Saved evaluation metrics
├─ requirements.txt
└─ app.py                       # Streamlit application entry point

Streamlit App Overview

The app loads the trained model from models/final_model.pkl and expects input features that match the columns and order of data/cleaned_X.csv.

• Single Prediction tab: Fill in patient details; the app builds the one-hot feature vector behind the scenes and predicts risk (and probability if available).
• Batch Prediction tab: Upload a CSV with the exact same header as data/cleaned_X.csv to get predictions for many rows at once. A header template download is provided in the UI.
• Documentation tab: Project overview, metrics (from results/evaluation_metrics.txt if present), and the exact feature column list.

Setup

• Recommended Python: 3.11 (Windows supported)
• Create/activate a virtual environment (optional but recommended)

Windows PowerShell example:

# From the project root
python -m venv .venv
. .venv\Scripts\Activate.ps1
pip install --upgrade pip
pip install -r requirements.txt

Run the App

From the project root:

streamlit run app.py

Streamlit will open a local URL in your browser. If it doesn’t, copy the printed URL into your browser.

Input Schema (very important)

The app derives the exact expected input columns from data/cleaned_X.csv. These include:

• Continuous: age, trestbps, chol, thalach, oldpeak, ca
• One-hot encoded groups (column names start with): sex_, cp_, restecg_, slope_, fbs_, exang_, thal_

For batch predictions, the uploaded CSV must have the identical header (names and order). Any extra columns will be ignored; any missing required columns will stop the run.

Troubleshooting

• UnpicklingError / invalid load key:

  • The model file may have been saved with joblib. The loader in app.py tries joblib.load first, then falls back to pickle.load.

  • If it still fails, ensure your environment’s scikit-learn version is compatible with the artifact that produced final_model.pkl. As a last resort, re-save the model in the current environment:

    import joblib
    joblib.dump(model, 'models/final_model.pkl')

• Model or data file not found:

  • Verify models/final_model.pkl and data/cleaned_X.csv exist and paths match the project layout.

• Feature mismatch for batch upload:

  • Use the header template download in the Batch tab and populate rows under that header.

Reproducibility Notes

• Keep final_model.pkl and cleaned_X.csv in sync with the training pipeline used in the notebooks.
• If you change preprocessing or feature engineering, regenerate both the features and the model.

License

This project is provided for educational and research purposes. Add a license file if you plan to distribute.