Scholarlink

<p>The dramatic growth of big data across commercial ecosystems has intensified the need for intelligent, privacy-preserving customer segmentation methods capable of generating actionable insights without compromising individual privacy rights. Traditional centralized machine learning approaches introduce substantial ethical, legal, and social risks including data privacy violations, algorithmic bias, and non-adherence to evolving regulatory frameworks such as GDPR (Aldweesh et al. 2024) (Yan and Hua 2024). This paper presents Federated Learning (FL) as a principled ethical framework for privacy-preserving customer segmentation in big data environments. Through systematic conceptual analysis and simulation-based experiments, we demonstrate how FL—augmented with Differential Privacy (DP), Secure Multi-Party Computation (SMPC), and Homomorphic Encryption (HE)—enables decentralized model training without centralizing raw personal data (Vanga 2026) (Hameed et al. 2025). The proposed Federated Differentiated Privacy Customer Segmentation (FDPCS) framework is validated through comparative simulation achieving competitive clustering quality (Silhouette Coefficient: 0.559 vs. 0.621 centralized) with significantly reduced privacy risk (membership inference attack accuracy: 51.1% vs. 73.4%). We further address key limitations including dataset preprocessing transparency, privacy budget justification ({ε} ∈ {0.5, 1.0}), scalability analysis, and fairness metrics (demographic parity, equal opportunity) to improve reproducibility and empirical grounding. The study contributes an interdisciplinary perspective bridging computer science, business intelligence, and data ethics.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Federated learning as an ethical framework for privacy-preserving customer segmentation in big data

  • 期刊论文

  • 发表日期：

    2026年5月4日

查看全文
  • Vivek Gondalia,
  • Disha H. Parekh,
  • Nehal Adhvaryu

摘要

The dramatic growth of big data across commercial ecosystems has intensified the need for intelligent, privacy-preserving customer segmentation methods capable of generating actionable insights without compromising individual privacy rights. Traditional centralized machine learning approaches introduce substantial ethical, legal, and social risks including data privacy violations, algorithmic bias, and non-adherence to evolving regulatory frameworks such as GDPR (Aldweesh et al. 2024) (Yan and Hua 2024). This paper presents Federated Learning (FL) as a principled ethical framework for privacy-preserving customer segmentation in big data environments. Through systematic conceptual analysis and simulation-based experiments, we demonstrate how FL—augmented with Differential Privacy (DP), Secure Multi-Party Computation (SMPC), and Homomorphic Encryption (HE)—enables decentralized model training without centralizing raw personal data (Vanga 2026) (Hameed et al. 2025). The proposed Federated Differentiated Privacy Customer Segmentation (FDPCS) framework is validated through comparative simulation achieving competitive clustering quality (Silhouette Coefficient: 0.559 vs. 0.621 centralized) with significantly reduced privacy risk (membership inference attack accuracy: 51.1% vs. 73.4%). We further address key limitations including dataset preprocessing transparency, privacy budget justification ({ε} ∈ {0.5, 1.0}), scalability analysis, and fairness metrics (demographic parity, equal opportunity) to improve reproducibility and empirical grounding. The study contributes an interdisciplinary perspective bridging computer science, business intelligence, and data ethics.