Inferring household size distribution and its association with the built environment using massive mobile phone data

Household size and its spatial distributions reflect not only the socioeconomic development in a city but also the rationality of urban resource allocation. Most existing studies rely heavily on census data to explore the potentially influential factors using methods such as macro-statistical analysis and socioeconomic analysis, of which the spatial resolutions and geographic scales are constrained. More importantly, the association between the household size distribution and the built environment is oversimplified or even neglected to some extent. In this work, we use massive mobile phone data combined with travel surveys of Beijing inhabitants' data (TSBI) to infer the household size and analyze the effect of spatial heterogeneity in a finer spatial resolution in Beijing, China. First, the machine learning method (i.e., support vector machine (SVM)) is applied to identify the household relationships of mobile users, and there are around 3.44 million households (families) with different sizes are obtained. Second, we analyze the spatial distribution patterns of household size and its association with built environmental indicators (e.g., public service density, public transportation density, etc.). The results exhibit a heterogeneous effect of the regional built environment on average household size (AHS). For instance, “commercial density” and “administrative density” show a negative impact on household size, while “public service density” and “public transportation density” show positive correlations with household size. As a complement to census data, mobile phone data can be used to obtain the household size in real-time. This paper provides quantified evidence for government departments to allocate facilities in a more targeted, balanced, and reasonable way according to the regional differences in household size, which would potentially support the sustainable urban development.