Key points are not available for this paper at this time.
Abstract Accurate demand forecasting remains a critical challenge in retail operations, where imprecise predictions lead to inventory overstocking, stockouts, and suboptimal pricing strategies. Conventional forecasting models frequently struggle to model the complex, nonlinear interactions between demand and a multitude of influencing factors. To address this limitation, the present work proposes a hybrid model combining the Grey Wolf Optimizer with an Extreme Learning Machine (GWO-ELM) for high-accuracy retail demand forecast prediction. The GWO algorithm is utilized to systematically determine the optimal input weights and biases of the ELM, overcoming the drawbacks of random initialization and enhancing generalization performance. The model is rigorously evaluated using 5-fold cross-validation and 20 independent runs on a real-world retail dataset, demonstrating superior predictive accuracy with a test coefficient of determination (R2) of 0.992 and significantly reduced error metrics compared to standard ELM and other metaheuristic variants. To ensure model transparency and actionable insights, SHAP (SHapley Additive exPlanations) analysis is integrated to interpret feature contributions, revealing that Units Sold, Price, and Competitor Pricing are the most influential predictors. The results confirm that the GWO-ELM framework delivers superior forecasting accuracy compared to existing benchmarks, while simultaneously generating interpretable outputs that support strategic decision-making in inventory management and dynamic pricing. This study advances the body of knowledge on intelligent forecasting systems by combining optimization, machine learning, and explainability into a cohesive and practical solution for retail analytics.
Feda et al. (Fri,) studied this question.