A Stochastic Optimization Model for Multi-Echelon Retail Inventory Planning Driven by Multi-Source Data Fusion Demand Forecasting Under Supply-Side Disruptions

• Abiodun Okunola

  Ladoke Akintola University Technology

  Author

Retail supply chains face compounding challenges from demand volatility and supply-side disruptions, yet most inventory planning models rely on single-source forecasts and static disruption assumptions. This study develops and validates a stochastic optimization framework that integrates multi-source data fusion demand forecasting with disruption-aware inventory planning for multi-echelon retail networks. The methodology combines a Cross-Modal Attention Transformer (CAMT) for demand prediction using point-of-sale transactions, social media signals, and weather data with a stochastic programming model that explicitly parameterizes supply disruption frequency and severity. Using a dataset of 10.3 million retail transactions and synthetic disruption scenarios calibrated to COVID-19-era supply chain disruptions, the framework achieved 89.4% forecast accuracy (MAPE = 10.6%), outperforming seasonal naive baselines by 28.3 percentage points and single-source deep learning models by 12.7 percentage points. The stochastic optimization reduced total inventory costs by 18.7% compared to deterministic planning under disruption scenarios, with the most significant improvements observed when disruption severity exceeded moderate levels. Feature importance analysis revealed that supplier lead time variance and social media sentiment volatility were the strongest predictors of optimal safety stock adjustments. The study contributes a replicable framework for integrating multimodal demand signals with disruption-aware inventory optimization, offering retail practitioners a data-driven pathway to enhance supply chain resilience while maintaining cost efficiency.

• pdf

Copyright (c) 2026 Abiodun Okunola (Author)

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