Threshold-Sensitive Business Analytics for Urban Retail Density: Interpretable Machine Learning Evidence from Multi-Source Geospatial Data

Abstract

Urban retail density is increasingly shaped by the interaction between local demand, facility networks, competitive pressure, street-level perception, and accessibility. Yet many business analytics models used for retail site assessment still assume linear responses and therefore understate the threshold behavior that governs where retail clusters become viable, saturated, or strategically fragile. This study develops a threshold-sensitive business analytics framework for urban retail density using interpretable machine learning and multi-source geospatial data. The analytical design integrates retail points of interest, population mobility, road-network accessibility, urban service facilities, competitive context, and street-view perception indicators into a 500-meter grid-level dataset. XGBoost, random forest, and ordinary least squares models are compared, and SHAP-based feature attribution, partial dependence analysis, and threshold interpretation are used to translate model outputs into business and planning insights. The results show that nonlinear machine learning improves predictive performance over linear baselines, with the strongest gains observed for small-format, light-asset retail. Urban function and competition explain the largest share of light-asset retail density, while human perception, accessibility, and facility synergy are more important for capital-intensive retail. Several variables display clear threshold regimes: moderate general-market density supports small-format clustering, high aesthetic perception is required before large-format density increases, and excessive mobility may weaken the stability needed for capital-intensive sites. The findings contribute to business data analytics by demonstrating how interpretable machine learning can move retail location analysis from correlation ranking toward threshold-aware decision support.

Keywords: Urban retail density; business analytics; interpretable machine learning; XGBoost; SHAP; street-view perception; geospatial data; threshold effects