FedAVL: Automated Vertical Federated Learning For Heterogeneous Healthcare Data

Abstract

The democratization of Artificial Intelligence (AI) increasingly depends on automated solutions that reduce human effort across the machine learning pipeline. While Automated Machine Learning (AutoML) has gained momentum in centralized environments, its direct application to Federated Learning (FL) remains limited, particularly in the underexplored area of Vertical Federated Learning (VFL). This paper proposes FedAVL, a novel framework that integrates AutoML capabilities into VFL by automating feature alignment, model selection, and hyperparameter optimization across clients with vertically partitioned data. Unlike existing frameworks such as AutoFL, which target horizontal FL, FedAVL addresses challenges unique to VFL including heterogeneous feature spaces, resource asymmetry, and secure feature alignment. Our complexity analysis demonstrates that FedAVL achieves comparable communication and computation overhead to conventional VFL while significantly reducing configuration complexity. Experimental results on three real-world healthcare datasets (Life Expectancy, Obesity, and COVID-19 X-rays) show that FedAVL reduces system setup time by up to 35% and improves predictive accuracy by 3-8% compared to manually tuned VFL and other baselines. These results highlight FedAVL has a scalable and resource-aware approach to democratizing privacy-preserving healthcare analytics.

Department(s)

Computer Science

Keywords and Phrases

Automated Machine Learning; Healthcare; Optimization; Vertical Federated Learning

Document Type

Article - Conference proceedings

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2026 Institute of Electrical and Electronics Engineers, All rights reserved.

Publication Date

01 Jan 2026

Link to Full Text

Share

 
COinS