FL-PSO: A Federated Learning approach with Particle Swarm Optimization for Brain Stroke Prediction

Abstract

Healthcare is one of the significant application areas of Cyber-Physical Systems, wherein massive amounts of sensors and other physical entities are interconnected to each other. Diagnosing and predicting diseases at an early stage is crucial for any healthcare application and machine-learning approaches are widely explored for the same. However, the conventional machine learning approaches can lead to the leakage of sensitive information pertaining to patients. In this study, our primary objective is to develop a machine learning based framework for early brain stroke prediction. Federated Learning (FL) is included in the framework to preserve the privacy of the patient’s data which is used as the basis for brain stroke prediction. The hyperparameters of FL are further optimized using Particle Swarm Optimization (PSO) to yield predictions with enhanced accuracy without compromising with data privacy. The experimental research showed that the suggested FL-PSO framework outperformed its competitors in terms of metrics like accuracy, validating the superiority of the framework.