Federated Evolutionary Feature Selection: A Framework for Scalable, Privacy-Preserving Machine Learning

Abstract

Feature selection plays a crucial role in enhancing the performance of machine learning models, especially in high-dimensional data environments. Traditional feature selection methods, often manual and domain-specific, face limitations in scalability and efficiency when applied to large datasets. Evolutionary Computation (EC) techniques, inspired by natural evolution, have emerged as powerful, automated alternatives capable of exploring and optimizing complex search spaces. Recent advances from 2017 to 2024 have significantly improved EC-based feature selection, introducing cutting-edge multi-objective optimization methods, hybrid algorithms, and more. These innovations have been applied across diverse fields such as gene expression analysis, symbolic regression, and high dimensional feature selection, leading to enhanced accuracy and efficiency. This review captures notable contributions to EC-driven feature selection, including particle swarm optimization (PSO), genetic programming, and innovative algorithms such as the Membership Weight Salp Swarm Algorithm (MWSSA) and Binary Bee Foraging. Additionally, this survey highlights emerging challenges, such as computational scalability, crucial for advancing the efficiency and effectiveness of EC-based feature selection. The review aims to provide insights into how these approaches have shaped the landscape of feature selection from 2017 to 2024. Apart from that, this paper explores the integration of Evolutionary Computation (EC) techniques with Federated Learning (FL) to address feature selection in privacy-constrained, distributed datasettings. By leveraging EC's global search capabilities and FL's decentralized architecture, the proposed approach minimizes communication costs while ensuring data privacy. Extensive experiments demonstrate the effectiveness of this method in improving classification accuracy, reducing feature subsets, and maintaining scalability across diverse datasets. The findings underscore the potential of FL + EC as a transformative framework for privacy-preserving feature selection in real-world applications."

Keywords: Evolutionary Computation, Feature selection, Multi-ObjectiveOptimization, Genetic Programming, Federated Learning