Demand-Weighted Spatial Clustering for Temporally Stable Cross-Dock Location Identification: Evidence from a Real Logistics Network

Background/Objectives: Selecting candidate cross-docking locations is a critical strategic decision in early-stage logistics network design. Conventional approaches often aggregate all demand data into a single planning snapshot, hiding the difference between consistently important locations and those attractive only under average conditions. This study proposes a demand-weighted spatial clustering framework for identifying temporally stable candidate cross-docking locations. Methods: The framework applies clustering to each monthly period independently and evaluates candidate centers through a cross-month stability analysis. K-Means++ is the main method, K-Medoids provides cross-validation, and DBSCAN serves as a density-based reference. Cluster counts are selected using a composite metric based on the Silhouette score, Davies–Bouldin index, and Calinski–Harabasz score. A stability score ranks candidate zones by combining frequency of appearance and demand weight. Results: On a real Turkish logistics dataset of 375,259 orders from 563 customers over 16 months, the framework produced 12 consensus candidate locations, eight appearing in all 16 months. A single-period baseline gave only nine candidates and missed four moderate-demand zones. A paired t-test showed the K-Means++ and K-Medoids difference was significant (p < 0.001). Conclusions: The framework offers a repeatable, data-driven approach to candidate-location screening as a structured input to subsequent sustainability-oriented network optimization models.