A Dynamic Multi-Source Data Fusion Framework for Real-Time Demand Forecasting and Automated Inventory Optimization in Omnichannel Retail Supply Chains

• Ada John

  ladoke Akintola university of technology

  Author

The integration of heterogeneous data streams in omnichannel retail environments presents both opportunities and challenges for demand forecasting and inventory optimization. This study develops and validates a dynamic multi-source data fusion framework that integrates point-of-sale transactions, warehouse management system data, e-commerce clickstream signals, and external indicators to enable real-time demand prediction and automated replenishment decisions. Drawing on a design-based research methodology combining retrospective analysis of historical retail data with prospective simulation, the framework employs a hybrid forecasting architecture that stacks Long Short-Term Memory networks, Gradient Boosting Machines, and Seasonal ARIMA models through a meta-learning layer. The proposed system was evaluated against conventional static forecasting approaches across a simulated omnichannel retail network. Results demonstrate that the multi-source fusion framework achieved 89.4% forecast accuracy (MAPE reduction of 34.2% compared to baseline methods), with statistically significant improvements in inventory turnover (p < 0.001) and reductions in stockout events of 42.8%. Feature importance analysis identified promotional calendars, clickstream conversion rates, and regional weather indices as the strongest external predictors. The framework offers a replicable architecture for retailers seeking to transition from periodic, single-source forecasting to continuous, multi-source demand sensing. Practical implications include a 31.6% reduction in safety stock requirements and a 28.4% decrease in expedited shipping costs. The study contributes a validated conceptual model linking data fusion maturity to operational performance in omnichannel supply chains

• pdf

Copyright (c) 2026 Ada John (Author)

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