Privacy-preserving federated learning for heart disease prediction using FedAvg and SHAP -based interpretability

Federated Learning (FL), which allows medical institutions to collaborate on training without disclosing raw patient data, offers a framework for cardiac disease prediction that protects privacy. FL reduces security threats and guarantees adherence to privacy laws, in contrast to conventional machine learning methods that depend on centralised data storage. FL preserves efficiency and data privacy by using the FedAvg approach to aggregate updates after training models locally. The model’s performance, which included an AUC of 0.88, 85% accuracy, 82% precision, and 88% recall, was on par with centralised methods. Important characteristics including age, cholesterol, and ECG readings were found to significantly influence predictions, which improved the interpretability of the model by providing SHAP-based justifications. This study shows that FL provides a workable Artifical Intelligence(AI) solution for diagnosing heart disease while maintaining privacy and legal compliance. In order to increase the clinical value of the model and boost confidence in AI-driven healthcare applications, future research will examine the integration of multi-modal data, scalability, and real-time updates.