The estimation of spatial distribution patterns of different socio-economic status (SES) groups by housing advertisement data and machine learning techniques: a case study in brooklyn, new york

Abstract

Poverty eradication has long been a central issue for sustainable development goals (SDGs), which draws attention to the issue of urban inequalities that can hinder regional economic development and increase unemployment and crime rates. It is critical to understand the local socio-economic distribution pattern for better urban policies and planning strategies. Traditional SES measurements are mainly based on census data and surveys, which are slowly updated and often fail to apply in the latest analysis. The SES inference methods using other data (e.g., satellite maps, nighttime lighting data) lack a theoretical basis and are of coarse resolution. The study takes advantage of the latest data (i.e., online housing advertisement data) and point of interests (POIs) to infer fine-grained block-group-level SES in Brooklyn through machine learning techniques. In addition, natural language processing (NLP) methods are used to derive twelve housing-related SES predictors. The results show that the speculative models and predictors are feasible, and the Global decision tree (GBDT) algorithm is the most efficient of the seven algorithms. The SES distribution map demonstrates a clear socio-economic stratification in Brooklyn. The rich are mainly concentrated in the western and northern areas with a high density of facilities. Based on the analysis of the local SES, three policy recommendations are proposed. First, for the inequitable distribution of facilities, additional investment should be made in the central and eastern regions. Second, a high level of greenery should be given priority in urban planning. Third, in terms of housing, disadvantaged groups should be given attention.