Abstract

Retail inventory systems increasingly operate under persistent supply disruptions arising from geopolitical shocks, climate events, transportation failures, and upstream capacity constraints. While machine learning has substantially improved short-term demand prediction accuracy, most state-of-the-art forecasting systems remain opaque, brittle under regime shifts, and poorly aligned with inventory decision making under uncertainty. Evidence across retail, healthcare, and public-sector systems suggests that predictive accuracy alone is insufficient to sustain operational performance when structural volatility is present (Hasan et al., 2021; Hasan et al., 2025).
This paper develops an explainable demand forecasting framework explicitly designed for disrupted retail supply chains. Drawing on theories of information asymmetry, organizational sensemaking, and robust operations, demand forecasting is conceptualized as a socio-technical decision system rather than a purely predictive task. The study advances a hybrid analytical framework that integrates multi-horizon machine-learning models with structured explainability mechanisms and disruption-aware latent state representations. Using large-scale U.S. retail transaction data augmented with supply shock indicators, the analysis demonstrates how explainability alters forecast trust, inventory responsiveness, and system stability under disruption.
Results show that explainable models achieve predictive accuracy comparable to black-box benchmarks while significantly improving forecast calibration, disruption detection, and downstream inventory performance. The contribution is threefold. First, the paper embeds explainability as a structural property of demand forecasting rather than a post hoc diagnostic. Second, it introduces a disruption-aware modeling architecture linking demand signals to supply-side constraints. Third, it provides empirically grounded guidance for inventory decision makers operating in volatile environments. The findings reposition explainable analytics as a necessary condition for resilient retail operations rather than a compliance-driven add-on.