Distributed and Collaborative Learning Approach for Stroke Prediction

In this paper, we focus on solving a binary classification problem for stroke prediction. The proposed approach is based on a decentralized and collaborative learning without data sharing among hospitals. This federated learning from decentralized electronic health records will provide a relevant framework for multi-institutional collaborations while maintaining data privacy for each participant. We focus on Artificial Neural Network classifier based on distributed medical data of patients. Simulation tests showed the good performances of the proposed approach, which achieves prediction accuracy of 92% in case of two centers and 95% in case of three centers.