Integrating spatial statistics and machine learning to identify relationships between e-commerce and distribution facilities in Texas, US

Abstract

This paper proposes a novel analytical framework that integrates spatial statistics and machine learning techniques to identify relationships between e-commerce and distribution facilities. The framework includes centrographic analysis, global and local spatial association measurements, and a recently popularized interpretable machine learning approach – gradient boosting decision trees (GBDT) – to analyze warehousing location choices. We apply this framework to ZIP Codes Business Patterns data from 2003 to 2016 in three large metropolitan areas in Texas, US (i.e., Dallas-Fort Worth, Austin, and Houston). The thematic maps reveal the spatial clustering of areas with higher e-commerce activity but lower logistics facility coverage. It is worth noting that we do not observe logistics sprawl in the study region. The GBDT results show that industrial activities and transportation network accessibility are key factors influencing warehousing location choices. We also find that the relationship between warehouses and e-commerce establishments is weaker in Houston, a major maritime gateway for goods entering and leaving, as compared to Dallas-Fort Worth and Austin. Implications for local freight transportation planners and decision-makers are discussed.