The death of neighborhood food outlets: Untangling spatial determinants of food retail environment vulnerability via interpretable machine learning

The global transformation of the retail environment has led to the widespread closure of small food outlets, threatening urban sustainability and dietary health. While extensive studies have explored the relationship between urban environments and food outlet closures, most of them do not differentiate between the closures of healthy and unhealthy food outlets. To address this gap, this study employs interpretable machine learning methods to examine the heterogeneity and differences in the spatial factors influencing healthy and unhealthy food outlet closures across 4328 neighborhoods in central Hangzhou, China, from 2019 to 2024. The findings reveal the following: (1) the median closure rate of healthy food outlets (47.2 %) was substantially higher than that of unhealthy ones (27.4 %), especially in suburban areas; (2) the impact of spatial variables on food outlet closures varied by neighborhood location; (3) a high supermarket density (>1.5 stores/km²) reduced the closure risk of healthy outlets, while a high density of small retail outlets (>200 stores/km²) mitigated closures of unhealthy outlets; (4) improved connectivity indicators, such as higher bus stop and sidewalk density, better protected healthy than unhealthy outlets. Based on these findings, we provide policy insights to foster a fairer, healthier, and more sustainable urban food retail environment.