Optimizing delivery systems within the e-retail context: a weighted self-organizing map for delivery region partitioning

Abstract

Today, e-tailing operations are well-established. However, managing dispersed same-day, next-day, or immediate deliveries remains a significant challenge. This necessitates refined vehicle routing and scheduling, which depends on efficient partitioning of the delivery regions. To tackle this, this paper develops a novel Weighted Self-Organizing Map Delivery Region Partitioning (WSOM-DRP) model that jointly generates delivery clusters and suggests optimal collection points for e-orders within each cluster. Using real data from a third-party logistics provider, our model is evaluated against alternative clustering methods (k-means, Ward hierarchical clustering, fuzzy c-means) using common clustering performance measures, travel distance and computation time. A comprehensive sensitivity analysis across varying cluster numbers confirms the model’s robustness, showing travel distance reduction of up to 36 % compared to the second-best method, particularly in high-density and high-traffic scenarios. Additionally, it yields significant improvements in clustering quality (e.g., a minimum of 15 % improvement in the silhouette index across scenarios) and an 18 % reduction in computation time compared to the next fastest benchmark. These findings highlight the practical value and adaptability of WSOM-DRP for optimizing delivery operations under diverse operational conditions and across different cluster granularities. The model also offers guidance on how to balance efficiency gains with operational complexity when selecting the number of clusters. By generating efficient delivery partitions and recommending optimal e-order collection locations during online checkout, our proposed WSOM-DRP model offers an e-commerce solution which is delivery efficient and cost-effective.