A Reproducible Privacy-Preserving Federated Learning Framework for CRM Decision Systems

Abstract

This repository contains the reproducibility materials for the FedCRM-DP study, a privacy-preserving cross-silo federated learning framework combining federated optimization, per-client record-level differential privacy, and SecAgg+ secure aggregation. The package contains source code, frozen experimental configurations, evaluation scripts, privacy-accounting artifacts, secure-aggregation diagnostics, final evaluation outputs, statistical analyses, and supporting documentation used to reproduce and audit the reported experiments on the Bank Marketing and Online Shoppers datasets. The study evaluates predictive performance, privacy–utility trade-offs, secure-aggregation behavior, and empirical membership-inference attacks under a frozen multi-seed experimental protocol. Please consult the included README and documentation for environment setup, execution instructions, experimental scope, and known limitations.