Abstract:

The dynamic growth of the e-commerce sector compels businesses to adopt multi-warehouse networks, creating a complex order allocation problem. Although Machine Learning (ML) models offer potential for automation, their "black box" nature constitutes a significant adoption barrier, eroding trust among managers. In response, this paper presents an integrated approach based on Explainable Artificial Intelligence (XAI), combining an XGBoost predictive model with the SHAP (SHapley Additive exPlanations) interpretation method. The system, trained on 24 months of real-world data, achieved an accuracy of 72.9%, significantly outperforming the baseline heuristic (67.1%). The key result is the demystification of the model’s decision-making process. SHAP analysis revealed that the model prioritizes order complexity and value (features min_font_size, all_products_value) over mere warehouse proximity. This work demonstrates that combining high performance with transparency creates a trustworthy tool capable of breaking down the adoption barriers for AI in logistics. Index Terms—Explainable Artificial Intelligence, XAI, ecommerce, logistics, order fulfillment, XGBoost, SHAP, optimization, supply chain management