Determinants of the localized behavior of individual food delivery choices

Online food delivery systems have grown rapidly and globally due to urbanization, busy lifestyles, flexible work-life integration, and widespread mobile and Internet use. Despite their benefits, such express delivery services cause issues such as traffic congestion and increasing carbon emissions. The distance traveled to deliver food in the city is directly related to the transportation cost and environmental impact of delivery. Thus, containing this distance while satisfying citizens' needs for diverse food choices is central for urban sustainability. This study introduces a novel localness index that accounts for cuisine type to measure the degree of locality in food orders, which can serve as a proxy for the carbon efficiency of food delivery. Then, it uses a tree-based machine learning model and Explainable AI methods to offer insights into how dozens of variables influence the localness index of food delivery choices. Using Dubai as a case study, the analysis shows that current food delivery behavior is highly non-local and carbon inefficient. Such non-local behavior is mainly driven by variables such as delivery fee, cuisine type, total wait duration, the number and variety of available food options nearby, restaurant ratings, menu item prices, incentives, and socioeconomic conditions of the neighborhood. Furthermore, these factors may influence the locality of food orders in different, sometimes opposing ways across various areas of Dubai City, warranting specific attention from both the food delivery platform and local policymakers in order to encourage ordering from local restaurants and reduce emissions.