A Hierarchical Density-Based Clustering Method Applied to Mixed-Mail in Austria

As the courier, express and parcel (CEP) market has grown rapidly in recent years, shipment packaging has also shifted from classic cuboid cardboards to mixed-mail, typically with flexible plastic or paper packaging. Despite being cost-effective and spaceefficient, the physical characteristics of mixed-mail items vary greatly, resulting in substantial difficulties when utilizing existing automated material handling technology in logistics distribution centers. Developing new material handling technologies that meet the requirements of mixed-mail is challenging due to the heterogeneity of the physical properties of mixedmail, making it difficult to find suitable specimens for testing. To address this issue, this study categorizes mixed-mail based on common combinations of physical characteristics using density-based cluster analysis. The physical characteristics of >400 mixed-mail items were recorded at an Austrian distribution center. The resulting dataset is of the mixed-variable type, meaning that it features both numerical and categorical variables. To homogenize the data for clustering, different methods are available. We compared four homogenization approaches using a benchmark study featuring simulated mixed-variable datasets with varying properties. The approach based on Gower’s distance in combination with the clustering algorithm Hierarchical Density-Based Spatial Clustering of
Applications with Noise (HDBSCAN) showed the best results over a wide range of different dataset properties. We then use this approach to cluster the Mixed-Mail dataset, resulting in two different clustering solutions based on different hyperparameter settings, with a total of six and eight clusters, respectively