Heart Stroke Prediction Using Federated Learning

Abstract

Heart attacks in youth have become a common issue, making it crucial to predict the chances of heart stroke in patients. However, the fragmented and private nature of healthcare data presents a significant challenge for producing reliable research results across populations. Federated Learning shows promise in connecting disparate healthcare data sources while also safeguarding privacy. To that end, this research aims to develop a Federated Learning algorithm that addresses the challenges and opportunities in healthcare by utilizing a central server to train a sharing globalized model while retaining sensitive data in local institutions where they originated. This approach is particularly important as electronic health records of diverse patient groups are owned by multiple institutions, making it difficult for hospitals to share such sensitive information. This proposed algorithm will contribute towards the creation of generalizable and efficient analytical techniques for predicting heart stroke risk in patients. This proposed algorithm produced better accuracies compared to traditional machine learning approaches. The study focused on working with numerical data sets and has the potential to contribute towards the creation of generalizable and efficient analytical techniques for predicting heart stroke risk in patients.