Mobility Networks as a Predictor of Socioeconomic Status in Urban Systems.

Modeling socioeconomic dynamics has always been an area of focus for urban scientists and policymakers, who aim to better understand and predict the well-being of local neighborhoods. Such models can inform decision-makers early on about expected neighborhood performance under normal conditions, as well as in response to considered interventions before official statistical data is collected. While features such as population and job density, employment characteristics, and other neighborhood variables have been studied and evaluated extensively, research on using the underlying networks of human interactions and urban structures is less common in modeling techniques. We propose using the structure of the local urban mobility network (weighted by commute flows among a city’s geographical units) as a signature of the neighborhood and as a source of features to model its socioeconomic quantities. The network structure is quantified through node embedding generated using a graph neural network representation learning model. In the proof-of-concept task of modeling the location’s median income and housing profile in two different cities, such network structure features provide a noticeable performance advantage compared to using only the other available social features. This work can thus inform researchers and stakeholders about the utility of mobility network structure in a complex urban system for modeling various quantities of interest.