Bridging Heterogeneous Data Silos: A Vertical Federated Learning Approach with Additive Secret Sharing for Cardiovascular Prediction

This study proposes a novel approach to overcome data silo challenges in healthcare by applying Vertical Federated Learning (VFL) with additive secret sharing for cardiovascular disease prediction. Using the UCI Heart Disease dataset, specifically the Cleveland and Hungarian datasets, our framework enables collaborative training across institutions with vertically partitioned patient data. Each institution trains a Convolutional Neural Network (CNN) on its localized data, contributing distinct patient features without sharing raw data. Secure aggregation is achieved through Secure Multi-Party Computation (SMPC) and additive secret sharing, ensuring compliance with privacy regulations while preserving model utility. Our global model demonstrates promising results in predicting cardiovascular risk, evaluated using metrics like F۱-score, ROC-AUC, and precision-recall, showcasing VFL's potential in healthcare applications.

کلیدواژه ها:

Vertical Federated Learning ، Additive Secret Sharing ، Cardiovascular Disease Prediction ، Healthcare Data Silos ، Convolutional Neural Network