Advanced Heart Attack Risk Prediction Using Stacked Hybrid Machine Learning

Aim:

         To design a privacy-preserving heart disease prediction model using Federated Learning (FL) that enables hospitals to collaboratively train machine learning models without sharing raw patient data.

Abstract:

       Heart disease is a major global health concern, and early risk prediction remains crucial for reducing mortality rates. Traditional machine learning approaches rely on centralized datasets, requiring hospitals to share sensitive patient information, which leads to significant privacy, security, and compliance challenges. Federated Learning (FL) offers an innovative solution by enabling multiple healthcare institutions to collaboratively train models without exposing raw data. In this study, a privacy-preserving federated framework is developed for heart disease prediction using a powerful ensemble of Random Forest, and XGBoost classifiers. Each participating institution independently trains a local model on its own patient dataset, and only encrypted model parameters are shared with the central server. The server aggregates these updates using algorithm to produce an improved global model. The ensemble-based federated approach enhances prediction accuracy, generalization, and robustness across diverse clinical environments. Experimental results show that FL achieves performance close to centralized learning while fully protecting patient confidentiality. The system is designed to comply with medical data regulations such as HIPAA and GDPR. This work demonstrates that Federated Learning enables secure, scalable, and collaborative AI for real-world healthcare applications.

Proposed System:

          The proposed system introduces a Federated Learning framework where hospitals train models locally on their private datasets. Instead of sharing patient data, they transmit only encrypted model updates to a central server. The server uses the Federated algorithm to combine updates and build a high-accuracy global model. An ensemble of Random Forest and XGBoost improves prediction robustness. This ensures data privacy while achieving accuracy comparable to centralized models. The system provides a scalable, secure, and collaborative solution for medical diagnosis.

Advantage:

• The system eliminates the need for raw patient data sharing between hospitals.
• It ensures complete privacy preservation throughout the entire model training process.
• The use of diverse datasets from multiple hospitals significantly improves overall prediction accuracy.
• The federated approach reduces model bias by learning from varied patient demographics and clinical conditions.
• It is highly scalable and can be deployed across numerous hospitals and medical organizations without compromising performance.