Federated Heart Disease Prediction
Predicting heart disease from clinical biomarkers using federated learning.
Links: GitHub repository
A federated learning study that predicts heart disease from clinical biomarkers without ever centralising patient records, so hospitals or clinics could collaborate on a shared model while keeping their data local.
Details
- Trains on a dataset of 918 subjects with 11 clinical features, including age, sex, chest pain type, resting blood pressure, cholesterol, fasting blood sugar, resting ECG, maximum heart rate, exercise-induced angina, oldpeak and ST slope, predicting the presence of heart disease.
- Uses a federated setup where each client trains locally and only model updates are shared and aggregated into a global model, preserving the privacy of sensitive medical data.
- Sits alongside my other federated-learning work as an exploration of privacy-preserving machine learning applied to healthcare biomarkers.