FMLCA: explainable and privacy-preserving federated machine learning classification algorithms for predicting heart...

Area of Science:

• Artificial Intelligence
• Medical Informatics
• Computational Biology

Background:

• Heart disease is a leading global cause of mortality.
• Machine Learning (ML) models are increasingly used for predicting Coronary Artery Disease (CAD).
• Federated Learning (FL) enables collaborative model training while preserving data privacy.

Purpose of the Study:

• To develop and evaluate Federated Machine Learning Classification Algorithms (FMLCA) for CAD prediction.
• To assess the performance of various ML models including Decision Tree, AdaBoost, KNN, Random Forest, and XGBoost.
• To incorporate privacy-preserving techniques like k-anonymity and utilize SHAP for model interpretability.

Main Methods:

• Implementation of FMLCA on a cloud computing platform.
• Application of k-anonymity for privacy preservation.
• Utilizing SHapley Additive exPlanations (SHAP) for feature importance analysis.
• Comparison of multiple ML classification algorithms for CAD prediction.

Main Results:

• The Random Forest (RF) model demonstrated superior performance among the evaluated algorithms.
• The RF model achieved an accuracy of 83.21% with privacy preservation and 84.49% without.
• SHAP analysis provided transparency by identifying key predictive features for CAD.

Conclusions:

• Cloud-based FMLCA offers efficient and accurate CAD prediction in distributed settings.
• The study highlights the importance of privacy and security in predictive healthcare.
• This approach advances predictive healthcare tools, supporting a patient-centric healthcare ecosystem.