FMLCA: explainable and privacy-preserving federated machine learning classification algorithms for predicting heart disease in patients

BACKGROUND: Heart disease is a global health concern that significantly contributes to worldwide mortality. Machine Learning (ML) models have emerged as a powerful tool for predicting Coronary Artery Disease (CAD), a type of heart disease, by utilizing clinical features for classification. Federated Learning (FL) offers a solution for collaborative training without sharing raw data, thus addressing privacy concerns.

METHODS: This study presents an innovative approach, Federated Machine Learning Classification Algorithms (FMLCA), which utilizes cloud computing, privacy preservation techniques, and ML classification algorithms, including Decision Tree (DT), Adaptive Boosting (AdaBoost), K-Nearest Neighbors (KNN), Random Forest (RF), and Extreme Gradient Boosting (XGBoost), to predict CAD. In addition, privacy preserving is considered through the k-anonymity technique, and SHapley Additive exPlanations (SHAP) technique was utilized to identify features important in the model decision-making process.

RESULTS: The proposed RF model, compared to other models, obtained better performance. This RF model achieved an accuracy of 83.21% with privacy preservation and 84.49% without it. Furthermore, the SHAP technique enhances transparency by attributing feature influences in predictions.

CONCLUSION: Implementing these models on a cloud platform results in efficient computational performance. This proposed approach represents a significant advancement in predictive healthcare tools, capable of accurately predicting CAD across distributed environments. By placing a strong emphasis on privacy and security, this approach underscores its importance and paves the way for a transformative healthcare ecosystem that centers on the needs of patients and healthcare providers.