Prediction and Optimization of Commodity Stagnation Risk for Cross-Border Multi-Overseas-Warehouse

Abstract

Aiming at the problems of inaccurate prediction of commodity stagnation risk, unbalanced inventory allocation, and high inventory holding cost and severe stagnation loss in the collaborative operation of multiple cross-border overseas warehouses, this paper constructs a multi-dimensional feature indicator system covering inventory attributes, supply chain operations, market sales, and overseas warehouse heterogeneity. With data optimization conducted via the 3σ criterion, Pearson correlation analysis and recursive feature elimination, a hybrid LightGBM model integrated with an attention mechanism is established. This study defines stagnation labels based on inventory turnover days in the next 30 days to accurately identify core risk factors. Taking accuracy, F1-score and other metrics as evaluation criteria, this paper compares the proposed model with multiple traditional machine learning models. On the basis of the predicted continuous stagnation risk scores, a multi-objective inventory optimization model is constructed to minimize inventory holding cost as well as stagnation and stockout losses. Combined with practical operational constraints, the improved NSGA-II algorithm is adopted for model solution to realize risk-oriented inventory allocation.