Machine Learning Pipeline for Heart Disease Prediction by Fares SultanMachine Learning Pipeline for Heart Disease Prediction by Fares Sultan

Machine Learning Pipeline for Heart Disease Prediction

Fares Sultan

Completed work

ML Engineer

pandas

Python

scikit-learn

Healthcare

Comprehensive Machine Learning Full Pipeline on Heart Disease UCI Dataset

An end-to-end machine learning project that analyzes, predicts, and visualizes heart disease risk using the Heart Disease UCI dataset. The pipeline covers data preprocessing, dimensionality reduction (PCA), feature selection, supervised classification, unsupervised clustering, hyperparameter tuning, and deployment through a Streamlit web app.

1. General Description

This project applies a full classical ML workflow to the Heart Disease UCI dataset:

Data preprocessing: cleaning, encoding, and scaling

Dimensionality reduction: PCA with explained-variance analysis

Feature selection: Random Forest importance, RFE, and Chi-Square test

Supervised learning: Logistic Regression, Decision Tree, Random Forest, SVM

Unsupervised learning: K-Means (Elbow Method) and Hierarchical Clustering

Hyperparameter tuning: GridSearchCV and RandomizedSearchCV

Deployment: a Streamlit UI for real-time prediction, deployable via Ngrok

2. Results Summary

Baseline model comparison (test set, 20% holdout):

Model Accuracy Precision Recall F1-score Logistic Regression 0.869 0.882 0.869 0.866 Decision Tree 0.721 0.721 0.721 0.720 Random Forest 0.853 0.860 0.853 0.850 SVM 0.853 0.870 0.853 0.849

After hyperparameter tuning:

Model Test Accuracy Random Forest (tuned) 0.836 SVM (tuned) 0.853

Best model deployed: SVM (tuned) — models/final_model.pkl

Clustering vs. actual labels: Adjusted Rand Index (K-Means, K=2) ≈ 0.30, indicating the natural K-Means clusters only partially align with the true diagnosis labels — expected, since clustering is unsupervised and unaware of the target.

Full logs and per-class metrics are in results/evaluation_metrics.txt. All generated charts are in results/plots/.

3. Dataset

Heart Disease UCI Dataset — 303 patient records, 13 clinical features (age, sex, chest pain type, resting blood pressure, cholesterol, etc.) plus a binary target (presence/absence of heart disease). A local copy is included at data/heart_disease.csv.

4. How to Run

Reproduce the full pipeline (plots + model)

This regenerates every plot in results/plots/, the metrics log in results/evaluation_metrics.txt, and the trained model in models/final_model.pkl.

Run step-by-step scripts

Run the scripts in notebooks/ in order (01 → 06) from inside that folder — each one saves its own plots/artifacts and some write intermediate CSVs the next script depends on:

These mirror run_pipeline.py stage by stage and work great as plain .py files in VS Code — just open the folder, open a terminal, and run each script.

Run the Streamlit app

The app loads models/final_model.pkl and lets you enter patient data for a real-time prediction. For exposing it publicly via Ngrok, see the deployment notes below.

Ngrok deployment (quick reference):

deployment/ngrok_setup.txt (kept locally, not committed) has the full walkthrough.

5. File Structure

Note: a local deployment/ngrok_setup.txt file also exists for personal reference but is excluded via .gitignore and not part of the pushed repo.

6. Tools Used

Python, Pandas, NumPy, Scikit-learn, Matplotlib, Seaborn, SciPy, Streamlit, Ngrok.

7. Limitations

Small dataset (303 records) — results should be interpreted with appropriate caution regarding generalization.

This is an educational project, not a medical diagnostic tool.

8. License

This project is licensed under the MIT License.

Like this project