Critical ethical challenges, such as data privacy, algorithmic governance, and equitable access, are increasingly relevant in a digital world where sensitive personal or community data are collected centrally or stored in the cloud. This study presents a comparative analysis of Centralized Learning and Federated Learning for privacy-preserving machine learning, using healthcare sleep disorder detection as a case study. The methodology employs ballistocardiography (BCG) signals acquired via non-invasive sensors, illustrating a replicable workflow for distributed, privacy-preserving analysis. Both approaches are evaluated using Key Performance Indicators (KPIs) to assess accuracy, privacy preservation, and computational efficiency. Experimental results show that the federated model achieves 77.98% testing accuracy, slightly surpassing the centralized model (75.71%), with only 18.27 s per training round and 2.91 s for testing, demonstrating fast convergence and competitive performance while preserving data privacy.

Enabling Privacy-Preserving Digital Well-Being Through Federated Learning

di Sivo D.;
2026-01-01

Abstract

Critical ethical challenges, such as data privacy, algorithmic governance, and equitable access, are increasingly relevant in a digital world where sensitive personal or community data are collected centrally or stored in the cloud. This study presents a comparative analysis of Centralized Learning and Federated Learning for privacy-preserving machine learning, using healthcare sleep disorder detection as a case study. The methodology employs ballistocardiography (BCG) signals acquired via non-invasive sensors, illustrating a replicable workflow for distributed, privacy-preserving analysis. Both approaches are evaluated using Key Performance Indicators (KPIs) to assess accuracy, privacy preservation, and computational efficiency. Experimental results show that the federated model achieves 77.98% testing accuracy, slightly surpassing the centralized model (75.71%), with only 18.27 s per training round and 2.91 s for testing, demonstrating fast convergence and competitive performance while preserving data privacy.
2026
Fusco, P.; Errico, P.; Di Sivo, D.; Venticinque, S.
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11383/2220611
 Attenzione

L'Ateneo sottopone a validazione solo i file PDF allegati

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? 0
social impact