Privacy-Preserving Federated Multi-Source Data Fusion Framework for Cross-Enterprise Retail Demand Forecasting via Differential Privacy and Secure Multi-Party Computation

• Abiodun Okunola

  Ladoke Akintola University Technology

  Author

Cross-enterprise retail demand forecasting requires the fusion of sensitive sales data from multiple suppliers, yet competitive concerns and regulatory constraints prevent direct data sharing. This study develops and validates a privacy-preserving federated multi-source data fusion framework integrating differential privacy (DP) and secure multi-party computation (SMPC) to enable collaborative demand forecasting without exposing proprietary data. Using a design-based research methodology with retrospective retail transaction data from 12 simulated enterprise participants spanning 36 months, the framework achieved 89.4% forecast accuracy (MAPE = 10.6%), outperforming isolated local models by 14.2 percentage points and a centralized non-private baseline by 2.8 percentage points while guaranteeing formal privacy protections. The DP noise injection resulted in only a 3.1% accuracy degradation relative to the non-private federated baseline, and SMPC computation overhead remained within 8.7% of unsecured aggregation. Feature importance analysis identified promotional events, seasonal indices, and cross-enterprise demand correlations as dominant predictors. The study concludes that DP-enhanced federated learning with SMPC aggregation offers a viable technical pathway for privacy-compliant collaborative forecasting in competitive retail ecosystems, providing a replicable framework for cross-enterprise data fusion under regulatory constraints.

• pdf

Copyright (c) 2026 Abiodun Okunola (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.