A Federated Learning Based Online Feature Selection Algorithm for Streaming Data

Abstract

In Federated Learning (FL), data privacy and communication overhead are considered significant challenges. In this paper, a dynamic feature selection mechanism is proposed integrated into the FL framework to address the complexity and communication cost issues arising from streaming data. Through experiments using the stochastic gradient descent (SGD) classifier on multiple streaming datasets, it is validated that this approach accelerates model training while improving predictive accuracy. This innovation not only reduces the data transmission between clients and the server but also enhances the efficiency of FL while ensuring data privacy. The findings of this study offer new insights for deploying FL in real-world applications, particularly in environments with limited computational resources and bandwidth.

Keywords: Federated Learning, Data Privacy, Dynamic Feature Selection, Communication Overhead, High-Dimensional Data