Scholarlink

<p>Big data analytics has enabled highly accurate customer segmentation, but centralized machine learning approaches raise serious concerns about privacy, algorithmic bias, and regulatory compliance. This paper introduces FedXAI-Seg, a novel framework that combines Federated Learning (FL) with SHAP-based Explainable AI (XAI) to perform RFM-based customer segmentation in a private, fair, and transparent manner. In FedXAI-Seg, local models are trained on distributed client data without sharing raw records; only differentially private model updates are exchanged, providing formal (ε = 1.0, δ = 10⁻⁵) privacy guarantees. Fairness is monitored using Demographic Parity Difference (DPD), Equalized Odds Difference (EOD), and Average Odds Difference (AOD). Experiments on the UCI Online Retail and Bank Marketing datasets show that FedXAI-Seg achieves a Silhouette Score of 0.631 (within 6.1% of the centralized baseline), reduces DPD by 71.1%, and lowers membership inference attack success rates by 30.9%. The framework is designed in compliance with GDPR Articles 5, 22, and 25, and the EU AI Act governance requirements.</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 for privacy-preserving customer segmentation: a fair and explainable framework for business intelligence

  • 期刊论文

  • 发表日期：

    2026年4月27日

查看全文
  • Kalpesh Popat,
  • Divyakant Meva,
  • Himanshu Maniar,
  • Hetal Modi,
  • Rutvi Shah

摘要

Big data analytics has enabled highly accurate customer segmentation, but centralized machine learning approaches raise serious concerns about privacy, algorithmic bias, and regulatory compliance. This paper introduces FedXAI-Seg, a novel framework that combines Federated Learning (FL) with SHAP-based Explainable AI (XAI) to perform RFM-based customer segmentation in a private, fair, and transparent manner. In FedXAI-Seg, local models are trained on distributed client data without sharing raw records; only differentially private model updates are exchanged, providing formal (ε = 1.0, δ = 10⁻⁵) privacy guarantees. Fairness is monitored using Demographic Parity Difference (DPD), Equalized Odds Difference (EOD), and Average Odds Difference (AOD). Experiments on the UCI Online Retail and Bank Marketing datasets show that FedXAI-Seg achieves a Silhouette Score of 0.631 (within 6.1% of the centralized baseline), reduces DPD by 71.1%, and lowers membership inference attack success rates by 30.9%. The framework is designed in compliance with GDPR Articles 5, 22, and 25, and the EU AI Act governance requirements.