A hybrid data prioritization and neural network framework for cost‑optimized supply chain management

The increasing complexity and volatility of modern supply chains necessitate intelligent decision-support systems that can efficiently process vast amounts of operational data while maintaining computational tractability. Traditional supply chain optimization models often fail to leverage the inherent structure of operational data, leading to suboptimal resource allocation and increased costs. Existing approaches in supply chain management (SCM) typically treat forecasting and optimization as separate stages, with data preprocessing methods, including computational sorting, serving merely as preliminary steps without any formal integration into the decision-making framework. This separation results in missed opportunities for improving operational efficiency and responsiveness. This study proposes a novel hybrid framework that formally integrates a data prioritization mechanism inspired by computational sorting principles with a neural network (NN)-based forecasting model within a unified cost-minimization mathematical programming framework. The prioritization mechanism generates a composite priority score (δij) derived from demand, cost, and inventory attributes, which is then embedded as an upper bound on assignment variables in the optimization model to guide resource allocation decisions. The framework is validated through comprehensive numerical experiments across 10 diverse supply chain scenarios, with each instance replicated 30 times to ensure statistical reliability. The primary scientific contribution lies in the mathematical formalization of the data prioritization process and its integration as operational constraints in the optimization model, rather than treating sorting as a mere preprocessing step. This formal linkage between data characteristics and optimization decisions distinguishes our approach from prior studies. Computational results, validated through Analysis of Variance (ANOVA) with F = 28.4 and p < 0.001, demonstrate that the proposed framework achieves superior performance. The Hybrid Framework reduces Mean Absolute Error (MAE) by 40.7% (from 178.50 to 105.80) compared to Single Sorting methods, and achieves a 12.1% reduction in average total cost compared to Data Science methods (from 14,528 to 12,769) and a 22.2% reduction compared to Single Sorting methods (from 16,407 to 12,769). The framework delivers a service level of 94.2%, significantly higher than the 86.6% achieved by Single Sorting and 89.6% by Data Science methods. Sensitivity analysis confirms robustness, maintaining service levels above 90% even under high demand variability (CV = 0.8), while scalability analysis demonstrates near-linear growth in computation time, with large-scale instances (50 × 20) solved in approximately 15 min (920 s). These findings underscore the practical potential of the proposed approach as a robust decision-support tool for strategic and tactical supply chain planning, offering managers a data-driven pathway to improve operational resilience and efficiency. Not applicable.